srameo / led

[ICCV 2023] Lighting Every Darkness in Two Pairs: A Calibration-Free Pipeline for RAW Denoising && [Arxiv 2023] Make Explicit Calibration Implicit: Calibrate Denoiser Instead of the Noise Model

Homepage | Paper | Google Drive | Baidu Cloud | 知乎 | MIPI Starting-Kit

📰News | 🔧Install | ✨Models Zoo | 📷Quick Demo | 🤖Benchmark | 🚧Contribute | 📜License | ❓FAQ

This repository contains the official implementation of the following papers:

Lighting Every Darkness in Two Pairs: A Calibration-Free Pipeline for RAW Denoising
Xin Jin^*, Jia-Wen Xiao^*, Ling-Hao Han, Chunle Guo^#, Ruixun Zhang, Xialei Liu, Chongyi Li
(* denotes equal contribution. # denotes the corresponding author.)
In ICCV 2023, [Paper Link], [Poster], [Slides], [Video]

Make Explicit Calibration Implicit: Calibrate Denoiser Instead of the Noise Model
Xin Jin, Jia-Wen Xiao, Ling-Hao Han, Chunle Guo^#, Xialei Liu, Chongyi Li, Ming-Ming Cheng^#
(# denotes corresponding authors.)
arxiv preprint, [Paper Link]

Comparaison with Calibration-Based Method

Some brief introduction on the process of calibration in [EN/CN].

LED is a Calibration-Free (or called implicit calibration) Pipeline for RAW Denoising (currently for extremely low-light conditions).

So tired of calibrating the noise model? Try our LED!
Achieveing SOTA performance in 2 paired data and training time less than 4mins!

More Teaser	More Teaser

• First of all, 🔧 Dependencies and Installation.
• For academic research, please refer to pretrained-models.md and 🤖 Training and Evaluation.
• For further development, please refer to 🚧 Further Development.
• For using LED on your own camera, please refer to ✨ Pretrained Models and 📷 Quick Demo.

📰 News

Future work can be found in todo.md.

• Jan 13, 2024: Release the starting-kit for MIPI@2024. Additionally, we release the pre-trained parameters of Restormer and NAFNet.
• Dec 27, 2023: Update an extension version of our ICCV 23 paper (Project Page/Paper).
• Dec 1-5, 2023: Add the related code/doc[EN/CN] from PR#14/PR#16, thanks to @HYX20011209
• Sep 27, 2023: Add the urls to our Poster, Slides, and Video.
• Aug 19, 2023: Release relevent files on Baidu Clould(pwd: iay5).

History

• Aug 15, 2023: For faster benchmark, we released the relevant files in commit fadffc7.
• Aug, 2023: We released a Chinese explanation of our paper on 知乎.
• Aug, 2023: Our code is publicly available!
• July, 2023: Our paper "Lighting Every Darkness in Two Pairs: A Calibration-Free Pipeline for RAW Denoising" has been accepted by ICCV 2023.

🔧 Dependencies and Installation

1. Clone and enter the repo:

   git clone https://github.com/Srameo/LED.git ICCV23-LED
   cd ICCV23-LED

2. Simply run the install.sh for installation! Or refer to install.md for more details.

   We use the customized rawpy package in ELD, if you don't want to use it or want to know more information, please move to install.md

   bash install.sh

3. Activate your env and start testing!

   conda activate LED-ICCV23

✨ Pretrained Models

If your requirement is for academic research and you would like to benchmark our method, please refer to pretrained-models.md, where we have a rich variety of models available across a diverse range of methods, training strategies, pre-training, and fine-tuning models.

We are currently dedicated to training an exceptionally capable network that can generalize well to various scenarios using only two data pairs! We will update this section once we achieve our goal. Stay tuned and look forward to it!
Or you can just use the following pretrained LED module for custumizing on your own cameras! (please follow the instruction in Quick Demo).

Method	Noise Model	Phase	Framework	Training Strategy	Additional Dgain (ratio)	Camera Model	Validation on	🔗 Download Links	Config File
LED	ELD (5 Virtual Cameras)	Pretrain	UNet	PMN	100-300	-	-	[Google Drive]	[options/LED/pretrain/MM22_PMN_Setting.yaml]
LED	ELD (5 Virtual Cameras)	Pretrain	UNet	ELD	100-300	-	-	[Google Drive]	[options/LED/pretrain/CVPR20_ELD_Setting.yaml]
LED	ELD (5 Virtual Cameras)	Pretrain	UNet	ELD	1-200	-	-	[Google Drive]	[options/LED/pretrain/CVPR20_ELD_Setting_Ratio1-200.yaml]
LED	ELD (5 Virtual Cameras)	Pretrain	Restormer	ELD	100-300	-	-	[Google Drive]	[options/LED/other_arch/Restormer/LED+Restormer_Pretrain.yaml]
LED	ELD (5 Virtual Cameras)	Pretrain	NAFNet	ELD	100-300	-	-	[Google Drive]	[options/LED/other_arch/NAFNet/LED+NAFNet_Pretrain.yaml]

📷 Quick Demo

Get Clean Images in the Dark!

We provide a script for testing your own RAW images in image_process.py.
You could run python scripts/image_process.py --help to get detailed information of this scripts.

If your camera model is one of {Sony A7S2, Nikon D850}, you can found our pretrained model in pretrained-models.md.

Notice that, if you wish to use the model from release v0.1.1, you need to add the -opt parameter: For NAFNet, add -opt options/base/network_g/nafnet.yaml. For Restormer, add -opt options/base/network_g/restormer.yaml.

usage: image_process.py [-h] -p PRETRAINED_NETWORK --data_path DATA_PATH [--save_path SAVE_PATH] [-opt NETWORK_OPTIONS] [--ratio RATIO] [--target_exposure TARGET_EXPOSURE] [--bps BPS] [--led]

optional arguments:
  -h, --help            show this help message and exit
  -p PRETRAINED_NETWORK, --pretrained_network PRETRAINED_NETWORK
                        the pretrained network path.
  --data_path DATA_PATH
                        the folder where contains only your raw images.
  --save_path SAVE_PATH
                        the folder where to save the processed images (in rgb), DEFAULT: 'inference/image_process'
  -opt NETWORK_OPTIONS, --network_options NETWORK_OPTIONS
                        the arch options of the pretrained network, DEFAULT: 'options/base/network_g/unet.yaml'
  --ratio RATIO, --dgain RATIO
                        the ratio/additional digital gain you would like to add on the image, DEFAULT: 1.0.
  --target_exposure TARGET_EXPOSURE
                        Target exposure, activate this will deactivate ratio.
  --bps BPS, --output_bps BPS
                        the bit depth for the output png file, DEFAULT: 16.
  --led                 if you are using a checkpoint fine-tuned by our led.

Fine-tune for Your Own Camera!

A detailed doc can be found in issue#8.

1. Collect noisy-clean image pairs for your camera model, please follow the insruction in demo.md.
2. Select a LED Pretrained model in our model zoo (based on the additional dgain you want to add on the image), and fine-tune it using your data!

   python scripts/cutomized_denoiser.py -t [TAG] \
                                        -p [PRETRAINED_LED_MODEL] \
                                        --dataroot your/path/to/the/pairs \
                                        --data_pair_list your/path/to/the/txt
   # Then the checkpoints can be found in experiments/[TAG]/models
   # If you are a seasoned user of BasicSR, you can use "--force_yml" to further fine-tune the details of the options.

3. Get ready and test your denoiser! (move to Get Clean Images in the Dark!).

🤖 Training and Evaluation

Please refer to benchmark.md to learn how to benchmark LED, how to train a new model from scratch.

🚧 Further Development

If you would like to develop/use LED in your projects, welcome to let us know. We will list your projects in this repository.
Also, we provide useful tools for your futher development, please refer to develop.md.

📖 Citation

If you find our repo useful for your research, please consider citing our paper:

@inproceedings{jiniccv23led,
    title={Lighting Every Darkness in Two Pairs: A Calibration-Free Pipeline for RAW Denoising},
    author={Jin, Xin and Xiao, Jia-Wen and Han, Ling-Hao and Guo, Chunle and Zhang, Ruixun and Liu, Xialei and Li, Chongyi},
    journal={Proceedings of the IEEE/CVF International Conference on Computer Vision},
    year={2023}
}

@inproceedings{jin2023make,
  title={Make Explict Calibration Implicit: "Calibrate" Denoiser Instead of The Noise Model},
  author={Jin, Xin and Xiao, Jia-Wen and Han, Ling-Hao and Guo, Chunle and Liu, Xialei and Li, Chongyi and Cheng, Ming-Ming},
  journal={Arxiv},
  year={2023}
}

📜 License

This code is licensed under the Creative Commons Attribution-NonCommercial 4.0 International for non-commercial use only. Please note that any commercial use of this code requires formal permission prior to use.

📮 Contact

For technical questions, please contact xjin[AT]mail.nankai.edu.cn and xiaojw[AT]mail.nankai.edu.cn.

For commercial licensing, please contact cmm[AT]nankai.edu.cn.

🤝 Acknowledgement

This repository borrows heavily from BasicSR, Learning-to-See-in-the-Dark and ELD.
We would like to extend heartfelt gratitude to Ms. Li Xinru for crafting the exquisite logo for our project.

We also thank all of our contributors.

led's Issues

關於訓練與測試在我的數據集上

您好，看完您的LED後我覺得非常的有興趣，想請問一下我手上有自己拍的一些數據集，我在晚上拍出來的raw影像。
我的這些影像是RGGB的並且像素值是在[0:1023]，如果是這樣的話有辦法使用您的LED做訓練與測試嗎?
如果可以的話我該從哪裡開始著手修改?

謝謝您!

关于噪声标定的曝光时间选择

您好，非常感谢你的开源工作！
我想问一下对于噪声标定数据的曝光时间选择有什么要求吗，不同曝光时间所采集的标定数据会对结果有影响吗？比如我在1/4000和1/10000分别进行标定，理论上结果是一致的吗？

为什么TL方差和K在对数形式上成线性关系了？

如题，这里一直不太懂 ，是有什么数学原理吗

请问您这边的rawpy包是什么版本的？为什么我这边会报错'rawpy._rawpy.RawPy' object has no attribute 'rgb_camera_matrix'

repnr_nafnet训练后推理 报错Unexpected key(s) in state_dict

你好，我在复现过程出现如下问题:
我使用repnr_nafnet.yaml 并修改了网络结构如下:
repnr_opt:
dont_convert_module:
- 'ending'
- 'dwconv2'
- 'downconv'
- 'upconv'
branch_num: 5
align_opts:
init_weight: 1.0
init_bias: 0.0
network_g:
type: NAFNet
img_channel: 4
width: 32
enc_blk_nums: [2, 2, 4, 8]
middle_blk_num: 2
dec_blk_nums: [2, 2, 2, 2]
训练完成之后，推理出现Unexpected key(s) in state_dict 错误，
Unexpected key(s) in state_dict: "intro.main_weight", "intro.main_bias", "intro.align_weights.0", "intro.align_weights.1", "intro.align_weights.2", "intro.align_weights.3", "intro.align_weights.4", "intro.align_weights.5", "intro.align_biases.0", "intro.align_biases.1", "intro.align_biases.2", "intro.align_biases.3", "intro.align_biases.4", "intro.align_biases.5", "encoders.0.0.conv1.main_weight", "encoders.0.0.conv1.main_bias", "encoders.0.0.conv1.align_weights.0", "encoders.0.0.conv1.align_weights.1", "encoders.0.0.conv1.align_weights.2", "encoders.0.0.conv1.align_weights.3", "encoders.0.0.conv1.align_weights.4", "encoders.0.0.conv1.align_weights.5", "encoders.0.0.conv1.align_biases.0", "encoders.0.0.conv1.align_biases.1", "encoders.0.0.conv1.align_biases.2", "encoders.0.0.conv1.align_biases.3", "encoders.0.0.conv1.align_biases.4", "encoders.0.0.conv1.align_biases.5", "encoders.0.0.conv3.main_weight", "encoders.0.0.conv3.main_bias", "encoders.0.0.conv3.align_weights.0", "encoders.0.0.conv3.align_weights.1", "encoders.0.0.conv3.align_weights.2", "encoders.0.0.conv3.align_weights.3", "encoders.0.0.conv3.align_weights.4", "encoders.0.0.conv3.align_weights.5", "encoders.0.0.conv3.align_biases.0", "encoders.0.0.conv3.align_biases.1", "encoders.0.0.conv3.align_biases.2", "encoders.0.0.conv3.align_biases.3", "encoders.0.0.conv3.align_biases.4", "encoders.0.0.conv3.align_biases.5", "encoders.0.0.sca.1.main_weight", "encoders.0.0.sca.1.main_bias", "encoders.0.0.sca.1.align_weights.0", "encoders.0.0.sca.1.align_weights.1", "encoders.0.0.sca.1.align_weights.2", "encoders.0.0.sca.1.align_weights.3", "encoders.0.0.sca.1.align_weights.4", "encoders.0.0.sca.1.align_weights.5", "encoders.0.0.sca.1.align_biases.0", "encoders.0.0.sca.1.align_biases.1", "encoders.0.0.sca.1.align_biases.2", "encoders.0.0.sca.1.align_biases.3", "encoders.0.0.sca.1.align_biases.4", "encoders.0.0.sca.1.align_biases.5", "encoders.0.0.conv4.main_weight", "encoders.0.0.conv4.main_bias", "encoders.0.0.conv4.align_weights.0", "encoders.0.0.conv4.align_weights.1", "encoders.0.0.conv4.align_weights.2", "encoders.0.0.conv4.align_weights.3 .....
我对比了net的key和模型的key的数量和名称不一致。
不修改网络错误也会同样的错误。请问如何处理？谢谢

关于raw数据的噪声可能会使信号低于bl,或者高于wl

你好，我看代码中处理前会进行归一化，im = (im - bl) / (wl - bl)。我有个问题，就是因为噪声会导致信号上下波动，raw数据我没有做clip，当信号低于bl,或者高于wl会对训练有什么影响。

尝试使用了LED + NAFNet 与 LED + Restormer 效果都不如Unet好

想问一下大家有无这样的情况 在作者给的LED+NAFNet 和 LED+Restormer 的预训练权重上进行微调 得到的结果 没有在原来的UNet上进行微调的效果好。想寻求一下大家的帮助～ network_g 用的是‘LED-main/options/base/network_g‘下的相应配置文件

多卡训练问题

您好！我在训练LED的时候出现的单张显卡显存不够的情况，请问如何设置多卡训练？

LED_Deploy_SID_SonyA7S2_CVPR20_Setting_Ratio100-300.pth 模型测试报错

用image_process或python scripts/benckmark.py
-t test
-p pretrained/network_g/LED_Deploy_SID_SonyA7S2_CVPR20_Setting_Ratio100-300.pth
--dataset SID SonyA7S2
加载LED_Deploy_SID_SonyA7S2_CVPR20_Setting_Ratio100-300.pth 报错，错误为Missing key(s) in state_dict: "conv1_1.weight", "conv1_1.bias", "conv1_2.weight", "conv1_2.bias", "conv2_1.weight", "conv2_1.bias", "conv2_2.weight", "conv2_2.bias", "conv3_1.weight", "conv3_1.bias", "conv3_2.weight" ·······，是不是得从参数化之后采用用？

请问这个方案支持文生3D视频么？

FYI，谢谢

Params ... will not be optimized. Training error?

I'm trying to train model using two pairs of RAW photos prepared according to https://github.com/Srameo/LED/blob/main/docs/demo.md

After training I'm running LED with the following command:

python scripts/image_process.py -p /ICCV23-LED/experiments/v0/models/net_g_latest.pth --data_path /input/ --save_path /output/

Resulting pictures looks clear from noise, with normal colors but very blurred.
May the problem be in a lot of warnings about parameters that will not be optimized?
If not, what should be changed?

Here is training log:

# python3 ./scripts/cutomized_denoiser.py -t "v0" -p pretrained/LED_Pretrain_None_None_CVPR20_Setting_Ratio1-200.pth --dataroot /input/Calibration/ --data_pair_list /input/Calibration/list
Disable distributed.
Path already exists. Rename it to /ICCV23-LED/tb_logger/v0_archived_20240122_130426
2024-01-22 13:04:26,952 INFO: 
     ______                   __   __                 __      __
    / ____/____   ____   ____/ /  / /   __  __ _____ / /__   / /
   / / __ / __ \ / __ \ / __  /  / /   / / / // ___// //_/  / /
  / /_/ // /_/ // /_/ // /_/ /  / /___/ /_/ // /__ / /<    /_/
  \____/ \____/ \____/ \____/  /_____/\____/ \___//_/|_|  (_)
    
Version Information: 
        LED: 0.1.1
        PyTorch: 1.13.1+cu117
        TorchVision: 0.14.1+cu117
2024-01-22 13:04:26,952 INFO: 
  val:[
    val_freq: 99999.0
    save_img: False
    suffix: None
    calculate_metric_in_batch: True
    illumination_correct: True
    metric_in_srgb: False
    metrics:[
      psnr:[
        type: calculate_psnr
        crop_border: 2
        test_y_channel: False
      ]
      ssim:[
        type: calculate_ssim
        crop_border: 2
        test_y_channel: False
      ]
    ]
  ]
  logger:[
    print_freq: 200
    save_checkpoint_freq: 99999.0
    use_tb_logger: True
    wandb: None
  ]
  datasets:[
    train:[
      num_worker_per_gpu: 8
      batch_size_per_gpu: 1
      dataset_enlarge_ratio: 99999
      prefetch_mode: None
      name: DemoFinetuneDataset
      type: FewshotPairedRAWDataset
      dataroot: /input/Calibration/
      which_meta: gt
      data_pair_list: /input/Calibration/list
      zero_clip: False
      use_hflip: True
      use_rot: True
      crop_size: 1024
      phase: train
      scale: 1
    ]
  ]
  train:[
    generalize_first: True
    pixel_opt:[
      type: L1Loss
      loss_weight: 1.0
      reduction: mean
    ]
    total_iter: 1500
    align_iter: 1000
    oomn_iter: 500
    warmup_iter: -1
    align_opt:[
      optim_g:[
        type: Adam
        lr: 0.0001
        weight_decay: 0
        betas: [0.9, 0.999]
      ]
      scheduler:[
        type: HandieLR
        milestones: [999999]
        lrs: [0]
      ]
    ]
    oomn_opt:[
      optim_g:[
        type: Adam
        lr: 1e-05
        weight_decay: 0
        betas: [0.9, 0.999]
      ]
      scheduler:[
        type: HandieLR
        milestones: [999999]
        lrs: [0]
      ]
    ]
  ]
  repnr_opt:[
    dont_convert_module: ['conv10_1']
    branch_num: 5
    align_opts:[
      init_weight: 1.0
      init_bias: 0.0
    ]
    aux_conv_opts:[
      bias: True
      init: zero_init_
    ]
  ]
  network_g:[
    type: UNetArch
    inchannels: 4
    outchannels: 4
    channels: 32
  ]
  base: ['options/base/network_g/repnr_unet.yaml', 'options/base/finetune/CVPR20_ELD.yaml', 'options/base/val_and_logger.yaml']
  name: v0
  model_type: LEDFinetuneModel
  scale: 1
  num_gpu: 1
  manual_seed: 2022
  path:[
    pretrain_network_g: pretrained/LED_Pretrain_None_None_CVPR20_Setting_Ratio1-200.pth
    strict_load_g: False
    resume_state: None
    experiments_root: /ICCV23-LED/experiments/v0
    models: /ICCV23-LED/experiments/v0/models
    training_states: /ICCV23-LED/experiments/v0/training_states
    log: /ICCV23-LED/experiments/v0
    visualization: /ICCV23-LED/experiments/v0/visualization
  ]
  dist: False
  rank: 0
  world_size: 1
  auto_resume: False
  is_train: True
  root_path: /ICCV23-LED

load lq metas in mem...: 100%|███████████████████████████████████████████████████| 2/2 [00:00<00:00,  2.29it/s]
load gt metas in mem...: 100%|███████████████████████████████████████████████████| 2/2 [00:00<00:00,  2.47it/s]
2024-01-22 13:04:28,777 INFO: Dataset [FewshotPairedRAWDataset] - DemoFinetuneDataset is built.
/opt/conda/envs/LED-ICCV23/lib/python3.8/site-packages/torch/utils/data/dataloader.py:554: UserWarning: This DataLoader will create 8 worker processes in total. Our suggested max number of worker in current system is 4, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.
  warnings.warn(_create_warning_msg(
2024-01-22 13:04:28,777 INFO: Training statistics:
        Number of train images: 2
        Dataset enlarge ratio: 99999
        Batch size per gpu: 1
        World size (gpu number): 1
        Require iter number per epoch: 199998
        Total epochs: 1; iters: 1500.
2024-01-22 13:04:28,836 INFO: Network [UNetArch] is created.
2024-01-22 13:04:28,836 INFO: Convert UNetArch into RepNRBase using kwargs:
OrderedDict([('dont_convert_module', ['conv10_1']), ('branch_num', 5), ('align_opts', OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])), ('aux_conv_opts', OrderedDict([('bias', True), ('init', 'zero_init_')]))])
2024-01-22 13:04:29,626 INFO: Network: RepNRBase, with parameters: 22,619,512
2024-01-22 13:04:29,626 INFO: RepNRBase: UNetArch(
  (conv1_1): RepNRConv2d(
    4, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (conv1_2): RepNRConv2d(
    32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (pool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv2_1): RepNRConv2d(
    32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (conv2_2): RepNRConv2d(
    64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (pool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv3_1): RepNRConv2d(
    64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (conv3_2): RepNRConv2d(
    128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (pool3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv4_1): RepNRConv2d(
    128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (conv4_2): RepNRConv2d(
    256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (pool4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv5_1): RepNRConv2d(
    256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (conv5_2): RepNRConv2d(
    512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (upv6): ConvTranspose2d(512, 256, kernel_size=(2, 2), stride=(2, 2))
  (conv6_1): RepNRConv2d(
    512, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (conv6_2): RepNRConv2d(
    256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (upv7): ConvTranspose2d(256, 128, kernel_size=(2, 2), stride=(2, 2))
  (conv7_1): RepNRConv2d(
    256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (conv7_2): RepNRConv2d(
    128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (upv8): ConvTranspose2d(128, 64, kernel_size=(2, 2), stride=(2, 2))
  (conv8_1): RepNRConv2d(
    128, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (conv8_2): RepNRConv2d(
    64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (upv9): ConvTranspose2d(64, 32, kernel_size=(2, 2), stride=(2, 2))
  (conv9_1): RepNRConv2d(
    64, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (conv9_2): RepNRConv2d(
    32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1, 1, 1), padding_mode=constant,
    branch_num=5, align_opts=OrderedDict([('init_weight', 1.0), ('init_bias', 0.0)])
    forward_type=reparameterize, aux_conv_opts=OrderedDict([('bias', True), ('init', 'zero_init_')])
  )
  (conv10_1): Conv2d(32, 4, kernel_size=(1, 1), stride=(1, 1))
)
2024-01-22 13:04:29,642 INFO: Loading RepNRBase model from pretrained/LED_Pretrain_None_None_CVPR20_Setting_Ratio1-200.pth, with param key: [params].
2024-01-22 13:04:29,660 WARNING: Current net - loaded net:
2024-01-22 13:04:29,660 WARNING:   conv1_1.aux_bias
2024-01-22 13:04:29,660 WARNING:   conv1_1.aux_weight
2024-01-22 13:04:29,660 WARNING:   conv1_2.aux_bias
2024-01-22 13:04:29,660 WARNING:   conv1_2.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv2_1.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv2_1.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv2_2.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv2_2.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv3_1.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv3_1.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv3_2.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv3_2.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv4_1.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv4_1.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv4_2.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv4_2.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv5_1.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv5_1.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv5_2.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv5_2.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv6_1.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv6_1.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv6_2.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv6_2.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv7_1.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv7_1.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv7_2.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv7_2.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv8_1.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv8_1.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv8_2.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv8_2.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv9_1.aux_bias
2024-01-22 13:04:29,661 WARNING:   conv9_1.aux_weight
2024-01-22 13:04:29,661 WARNING:   conv9_2.aux_bias
2024-01-22 13:04:29,662 WARNING:   conv9_2.aux_weight
2024-01-22 13:04:29,662 WARNING: Loaded net - current net:
2024-01-22 13:04:29,683 INFO: Loss [L1Loss] is created.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv1_1.weight will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv1_1.bias will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv1_2.weight will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv1_2.bias will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv2_1.weight will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv2_1.bias will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv2_2.weight will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv2_2.bias will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv3_1.weight will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv3_1.bias will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv3_2.weight will not be optimized.
2024-01-22 13:04:29,683 WARNING: Params base_module.conv3_2.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv4_1.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv4_1.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv4_2.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv4_2.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv5_1.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv5_1.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv5_2.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv5_2.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.upv6.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.upv6.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv6_1.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv6_1.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv6_2.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv6_2.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.upv7.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.upv7.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv7_1.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv7_1.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv7_2.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv7_2.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.upv8.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.upv8.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv8_1.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv8_1.bias will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv8_2.weight will not be optimized.
2024-01-22 13:04:29,684 WARNING: Params base_module.conv8_2.bias will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params base_module.upv9.weight will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params base_module.upv9.bias will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params base_module.conv9_1.weight will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params base_module.conv9_1.bias will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params base_module.conv9_2.weight will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params base_module.conv9_2.bias will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params base_module.conv10_1.weight will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params base_module.conv10_1.bias will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.main_weight will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.main_bias will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.align_weights.0 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.align_weights.1 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.align_weights.2 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.align_weights.3 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.align_weights.4 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.align_biases.0 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.align_biases.1 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.align_biases.2 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.align_biases.3 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_1.align_biases.4 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_2.main_weight will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_2.main_bias will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_2.align_weights.0 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_2.align_weights.1 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_2.align_weights.2 will not be optimized.
2024-01-22 13:04:29,685 WARNING: Params repnr_module.conv1_2.align_weights.3 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv1_2.align_weights.4 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv1_2.align_biases.0 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv1_2.align_biases.1 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv1_2.align_biases.2 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv1_2.align_biases.3 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv1_2.align_biases.4 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.main_weight will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.main_bias will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.align_weights.0 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.align_weights.1 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.align_weights.2 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.align_weights.3 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.align_weights.4 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.align_biases.0 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.align_biases.1 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.align_biases.2 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.align_biases.3 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_1.align_biases.4 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_2.main_weight will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_2.main_bias will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_2.align_weights.0 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_2.align_weights.1 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_2.align_weights.2 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_2.align_weights.3 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_2.align_weights.4 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_2.align_biases.0 will not be optimized.
2024-01-22 13:04:29,686 WARNING: Params repnr_module.conv2_2.align_biases.1 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv2_2.align_biases.2 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv2_2.align_biases.3 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv2_2.align_biases.4 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.main_weight will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.main_bias will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.align_weights.0 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.align_weights.1 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.align_weights.2 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.align_weights.3 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.align_weights.4 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.align_biases.0 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.align_biases.1 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.align_biases.2 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.align_biases.3 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_1.align_biases.4 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.main_weight will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.main_bias will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.align_weights.0 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.align_weights.1 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.align_weights.2 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.align_weights.3 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.align_weights.4 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.align_biases.0 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.align_biases.1 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.align_biases.2 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.align_biases.3 will not be optimized.
2024-01-22 13:04:29,687 WARNING: Params repnr_module.conv3_2.align_biases.4 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.main_weight will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.main_bias will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.align_weights.0 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.align_weights.1 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.align_weights.2 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.align_weights.3 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.align_weights.4 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.align_biases.0 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.align_biases.1 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.align_biases.2 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.align_biases.3 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_1.align_biases.4 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.main_weight will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.main_bias will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.align_weights.0 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.align_weights.1 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.align_weights.2 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.align_weights.3 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.align_weights.4 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.align_biases.0 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.align_biases.1 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.align_biases.2 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.align_biases.3 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv4_2.align_biases.4 will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv5_1.main_weight will not be optimized.
2024-01-22 13:04:29,688 WARNING: Params repnr_module.conv5_1.main_bias will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_1.align_weights.0 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_1.align_weights.1 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_1.align_weights.2 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_1.align_weights.3 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_1.align_weights.4 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_1.align_biases.0 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_1.align_biases.1 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_1.align_biases.2 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_1.align_biases.3 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_1.align_biases.4 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.main_weight will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.main_bias will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.align_weights.0 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.align_weights.1 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.align_weights.2 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.align_weights.3 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.align_weights.4 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.align_biases.0 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.align_biases.1 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.align_biases.2 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.align_biases.3 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv5_2.align_biases.4 will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.upv6.weight will not be optimized, though it requires grad!
2024-01-22 13:04:29,689 WARNING: Params repnr_module.upv6.bias will not be optimized, though it requires grad!
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv6_1.main_weight will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv6_1.main_bias will not be optimized.
2024-01-22 13:04:29,689 WARNING: Params repnr_module.conv6_1.align_weights.0 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_1.align_weights.1 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_1.align_weights.2 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_1.align_weights.3 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_1.align_weights.4 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_1.align_biases.0 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_1.align_biases.1 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_1.align_biases.2 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_1.align_biases.3 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_1.align_biases.4 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.main_weight will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.main_bias will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.align_weights.0 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.align_weights.1 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.align_weights.2 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.align_weights.3 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.align_weights.4 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.align_biases.0 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.align_biases.1 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.align_biases.2 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.align_biases.3 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv6_2.align_biases.4 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.upv7.weight will not be optimized, though it requires grad!
2024-01-22 13:04:29,690 WARNING: Params repnr_module.upv7.bias will not be optimized, though it requires grad!
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv7_1.main_weight will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv7_1.main_bias will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv7_1.align_weights.0 will not be optimized.
2024-01-22 13:04:29,690 WARNING: Params repnr_module.conv7_1.align_weights.1 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_1.align_weights.2 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_1.align_weights.3 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_1.align_weights.4 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_1.align_biases.0 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_1.align_biases.1 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_1.align_biases.2 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_1.align_biases.3 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_1.align_biases.4 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.main_weight will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.main_bias will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.align_weights.0 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.align_weights.1 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.align_weights.2 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.align_weights.3 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.align_weights.4 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.align_biases.0 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.align_biases.1 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.align_biases.2 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.align_biases.3 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv7_2.align_biases.4 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.upv8.weight will not be optimized, though it requires grad!
2024-01-22 13:04:29,691 WARNING: Params repnr_module.upv8.bias will not be optimized, though it requires grad!
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv8_1.main_weight will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv8_1.main_bias will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv8_1.align_weights.0 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv8_1.align_weights.1 will not be optimized.
2024-01-22 13:04:29,691 WARNING: Params repnr_module.conv8_1.align_weights.2 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_1.align_weights.3 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_1.align_weights.4 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_1.align_biases.0 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_1.align_biases.1 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_1.align_biases.2 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_1.align_biases.3 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_1.align_biases.4 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.main_weight will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.main_bias will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.align_weights.0 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.align_weights.1 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.align_weights.2 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.align_weights.3 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.align_weights.4 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.align_biases.0 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.align_biases.1 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.align_biases.2 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.align_biases.3 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv8_2.align_biases.4 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.upv9.weight will not be optimized, though it requires grad!
2024-01-22 13:04:29,692 WARNING: Params repnr_module.upv9.bias will not be optimized, though it requires grad!
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv9_1.main_weight will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv9_1.main_bias will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv9_1.align_weights.0 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv9_1.align_weights.1 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv9_1.align_weights.2 will not be optimized.
2024-01-22 13:04:29,692 WARNING: Params repnr_module.conv9_1.align_weights.3 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_1.align_weights.4 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_1.align_biases.0 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_1.align_biases.1 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_1.align_biases.2 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_1.align_biases.3 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_1.align_biases.4 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.main_weight will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.main_bias will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.align_weights.0 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.align_weights.1 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.align_weights.2 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.align_weights.3 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.align_weights.4 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.align_biases.0 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.align_biases.1 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.align_biases.2 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.align_biases.3 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv9_2.align_biases.4 will not be optimized.
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv10_1.weight will not be optimized, though it requires grad!
2024-01-22 13:04:29,693 WARNING: Params repnr_module.conv10_1.bias will not be optimized, though it requires grad!
2024-01-22 13:04:29,694 INFO: Model [LEDFinetuneModel] is created.
2024-01-22 13:04:29,879 INFO: Start training from epoch: 0, iter: 0
2024-01-22 13:05:38,652 INFO: [v0..][epoch:  0, iter:     200, lr:(1.000e-04,)] [eta: 0:06:33, time (data): 0.344 (0.010)] l_pix: 1.1457e-02 
2024-01-22 13:06:40,049 INFO: [v0..][epoch:  0, iter:     400, lr:(1.000e-04,)] [eta: 0:05:34, time (data): 0.307 (0.007)] l_pix: 1.3481e-02 
2024-01-22 13:07:41,510 INFO: [v0..][epoch:  0, iter:     600, lr:(1.000e-04,)] [eta: 0:04:34, time (data): 0.307 (0.007)] l_pix: 9.3888e-03 
2024-01-22 13:08:42,970 INFO: [v0..][epoch:  0, iter:     800, lr:(1.000e-04,)] [eta: 0:03:33, time (data): 0.307 (0.006)] l_pix: 1.1264e-02 
2024-01-22 13:09:44,459 INFO: [v0..][epoch:  0, iter:   1,000, lr:(1.000e-04,)] [eta: 0:02:32, time (data): 0.307 (0.006)] l_pix: 1.2119e-02 
2024-01-22 13:09:44,475 INFO: Switch to optimize oomn branch....
2024-01-22 13:10:46,266 INFO: [v0..][epoch:  0, iter:   1,200, lr:(1.000e-05,)] [eta: 0:01:31, time (data): 0.309 (0.006)] l_pix: 1.0544e-02 
2024-01-22 13:11:48,080 INFO: [v0..][epoch:  0, iter:   1,400, lr:(1.000e-05,)] [eta: 0:00:30, time (data): 0.309 (0.006)] l_pix: 7.9759e-03 
2024-01-22 13:12:19,472 INFO: End of training. Time consumed: 0:07:49
2024-01-22 13:12:19,473 INFO: Save the latest model.

关于测试结果的疑问

我自己拍了一些raw图数据集，bayer模式是BGGR，也按照作者您提到的数据预处理对数据做了处理，但测试的结果偏绿，请问下这可能是什么原因导致的呢？

原gt图

lq图

如果想要 noise generator 来合成低光的图像，这里的ratio 和 scale 如何设置呢

关于noise_generatorg,单独运行合成噪声时遇到的问题

用您的噪声标定代码标定自己相机的噪声后，生成了一些参数k,row,tl,gauss四个参数，
1，可是在noise_generator中需要用lambda参数这个是从哪里获得的？
2，noise_generator中有两种合成方式一个是需要标定的一个是虚拟的，我有自己的相机用就是用需要标定的对吧？
3，在标定噪声对函数中的scale，和ratio 是以怎样的数据类型输入的？scale不是一个固定值么？
期待您的回复

_pickle.UnpicklingError: invalid load key

I was running clean ubuntu 20.04 in docker. After installing gcc, make, conda, etc.. I have downloaded LED repo following README.md instructions and run install.sh. After that I was downloaded pth-model, activated conda environment and run test script, which is failed:

(LED-ICCV23) root@host:/ICCV23-LED# python3 /ICCV23-LED/scripts/benckmark.py -p /ICCV23-LED/pretrained/SID_Deploy_SID_SonyA7S2_CVPR20_Setting_Ratio100-300.pth --dataset SID SonyA7S2 -t test
Disable distributed.
Path already exists. Rename it to /ICCV23-LED/results/test_archived_20231213_115820
2023-12-13 11:58:20,049 INFO:
     ______                   __   __                 __      __
    / ____/____   ____   ____/ /  / /   __  __ _____ / /__   / /
   / / __ / __ \ / __ \ / __  /  / /   / / / // ___// //_/  / /
  / /_/ // /_/ // /_/ // /_/ /  / /___/ /_/ // /__ / /<    /_/
  \____/ \____/ \____/ \____/  /_____/\____/ \___//_/|_|  (_)

Version Information:
        LED: 0.0.1
        PyTorch: 1.13.1+cu117
        TorchVision: 0.14.1+cu117
2023-12-13 11:58:20,049 INFO:
  val:[
    val_freq: 1288.0
    save_img: False
    suffix: None
    calculate_metric_in_batch: True
    illumination_correct: True
    metric_in_srgb: False
    metrics:[
      psnr:[
        type: calculate_psnr
        crop_border: 2
        test_y_channel: False
      ]
      ssim:[
        type: calculate_ssim
        crop_border: 2
        test_y_channel: False
      ]
    ]
  ]
  logger:[
    print_freq: 200
    save_checkpoint_freq: 1288.0
    use_tb_logger: True
    wandb: None
  ]
  datasets:[
    val_100:[
      name: SIDSonyA7S2Paired100
      type: PairedRAWDataset
      dataroot: datasets/ICCV23-LED/Sony_npy
      postfix: npz
      which_meta: gt
      data_pair_list: datasets/txtfiles/SID/SonyA7S2/valid_100.txt
      zero_clip: False
      crop_size: None
      load_in_mem: False
      phase: val
      scale: 1
    ]
    val_250:[
      name: SIDSonyA7S2Paired250
      type: PairedRAWDataset
      dataroot: datasets/ICCV23-LED/Sony_npy
      postfix: npz
      which_meta: gt
      data_pair_list: datasets/txtfiles/SID/SonyA7S2/valid_250.txt
      zero_clip: False
      crop_size: None
      load_in_mem: False
      phase: val
      scale: 1
    ]
    val_300:[
      name: SIDSonyA7S2Paired300
      type: PairedRAWDataset
      dataroot: datasets/ICCV23-LED/Sony_npy
      postfix: npz
      which_meta: gt
      data_pair_list: datasets/txtfiles/SID/SonyA7S2/valid_300.txt
      zero_clip: False
      crop_size: None
      load_in_mem: False
      phase: val
      scale: 1
    ]
  ]
  network_g:[
    type: UNetArch
    inchannels: 4
    outchannels: 4
    channels: 32
  ]
  base: ['options/base/network_g/unet.yaml', 'options/base/dataset/test/SID_SonyA7S2_val_split.yaml', 'options/base/val_and_logger.yaml']
  name: test
  model_type: RAWImageDenoisingModel
  scale: 1
  num_gpu: 1
  manual_seed: 2022
  path:[
    pretrain_network_g: /ICCV23-LED/pretrained/SID_Deploy_SID_SonyA7S2_CVPR20_Setting_Ratio100-300.pth
    strict_load_g: True
    resume_state: None
    CRF: datasets/EMoR
    results_root: /ICCV23-LED/results/test
    log: /ICCV23-LED/results/test
    visualization: /ICCV23-LED/results/test/visualization
  ]
  dist: False
  rank: 0
  world_size: 1
  auto_resume: False
  is_train: False

2023-12-13 11:58:20,049 INFO: Dataset [PairedRAWDataset] - SIDSonyA7S2Paired100 is built.
2023-12-13 11:58:20,050 INFO: Number of test images in SIDSonyA7S2Paired100: 29
2023-12-13 11:58:20,050 INFO: Dataset [PairedRAWDataset] - SIDSonyA7S2Paired250 is built.
2023-12-13 11:58:20,050 INFO: Number of test images in SIDSonyA7S2Paired250: 29
2023-12-13 11:58:20,051 INFO: Dataset [PairedRAWDataset] - SIDSonyA7S2Paired300 is built.
2023-12-13 11:58:20,051 INFO: Number of test images in SIDSonyA7S2Paired300: 35
2023-12-13 11:58:20,143 INFO: Network [UNetArch] is created.
2023-12-13 11:58:21,088 INFO: Network: UNetArch, with parameters: 7,760,484
2023-12-13 11:58:21,088 INFO: UNetArch(
  (conv1_1): Conv2d(4, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv1_2): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (pool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv2_1): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv2_2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (pool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv3_1): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv3_2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (pool3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv4_1): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv4_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (pool4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv5_1): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv5_2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (upv6): ConvTranspose2d(512, 256, kernel_size=(2, 2), stride=(2, 2))
  (conv6_1): Conv2d(512, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv6_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (upv7): ConvTranspose2d(256, 128, kernel_size=(2, 2), stride=(2, 2))
  (conv7_1): Conv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv7_2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (upv8): ConvTranspose2d(128, 64, kernel_size=(2, 2), stride=(2, 2))
  (conv8_1): Conv2d(128, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv8_2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (upv9): ConvTranspose2d(64, 32, kernel_size=(2, 2), stride=(2, 2))
  (conv9_1): Conv2d(64, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv9_2): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (conv10_1): Conv2d(32, 4, kernel_size=(1, 1), stride=(1, 1))
)
Traceback (most recent call last):
  File "led/test.py", line 45, in <module>
    test_pipeline(root_path)
  File "led/test.py", line 35, in test_pipeline
    model = build_model(opt)
  File "/ICCV23-LED/led/models/__init__.py", line 26, in build_model
    model = MODEL_REGISTRY.get(opt['model_type'])(opt)
  File "/ICCV23-LED/led/models/raw_image_denoising_model.py", line 56, in __init__
    self.load_network(self.net_g, load_path, self.opt['path'].get('strict_load_g', True), param_key)
  File "/ICCV23-LED/led/models/base_model.py", line 305, in load_network
    load_net = torch.load(load_path, map_location=lambda storage, loc: storage)
  File "/opt/conda/envs/LED-ICCV23/lib/python3.8/site-packages/torch/serialization.py", line 795, in load
    return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
  File "/opt/conda/envs/LED-ICCV23/lib/python3.8/site-packages/torch/serialization.py", line 1002, in _legacy_load
    magic_number = pickle_module.load(f, **pickle_load_args)
_pickle.UnpicklingError: invalid load key, '<'.

How to fix this?

Inquiry Regarding Performance Discrepancy between LED and DNF Algorithms

Dear @Srameo,

I am writing to you as I have come across your remarkable works, specifically "LED: Lighting Every Darkness in Two Pairs: A Calibration-Free Pipeline for RAW Denoising" and "DNF: Decouple and Feedback Network for Seeing in the Dark," and I am intrigued by the significant performance difference observed between the two algorithms.

To illustrate my query, I would like to reference the Sony split dataset from SID. In "LED's" Table 1, the LED algorithm achieves an impressive PSNR of 41.9842 and SSIM of 0.9539 for a 100x exposure. However, in "DNF's" Table 1, the DNF algorithm achieves a comparatively lower PSNR of 30.62 and SSIM of 0.797.

This stark contrast in performance has piqued my curiosity, and I am interested in understanding the factors that contribute to such a pronounced gap. It seems that the two algorithms might be tailored to address distinct tasks—one focusing on denoising exclusively, while the other perhaps incorporates a combination of Low-Light Image Enhancement (LLIE) and denoising.

Could you kindly shed some light on the underlying objectives and design philosophies of the LED and DNF algorithms? This clarification would greatly assist in deepening my comprehension of the nuances in your research.

Thank you for your time and consideration. I am genuinely fascinated by the advancements you have made in this field, and I am looking forward to enhancing my understanding of your work. Your expertise will undoubtedly contribute to my knowledge expansion.

I eagerly anticipate your response and thank you in advance for your assistance.

Best regards,

It seems something wrong with `python scripts/download_gdrive.py`

The current script itself relies on rawpy:

KeyError: 'network_g'

请问这个是怎么回事呢？ 看起来是yaml 不能读子yaml.

Please clarify fine-tuning instructions.

I am trying to use LED for my mobile phone camera.

Could you please clarify some questions about the documentation: https://github.com/Srameo/LED/blob/main/docs/demo.md

1. How to choose best Pretrain checkpoint for the further training depending on user case?

2. When preparing pairs, noisy image should be first in pair, right? If so, could you highlight this fact in the document?

3. Could you note in the document, that the cutomized_denoiser.py script will create experiments/[TAG] directory itself?

4. What recommendations concerning optimal fine-tuning pairs set? I.e. , is there any rules besides constant exposure value ratio? May be photo should contain clean vivid colors, one pair should be underexposed (to get "clear" noise) and one with normal exposition or so on?

p.s. For the mobile phone there is an app MotionCam , that allows create RAW photos with manual ISO and Exposition. Free version is enough for training purposes. The only problem with phone is attaching wireless pointer devices to avoid phone moving during preparing photos.

p.p.s. To control RAW exposition the following command can be used: exiftool *dng | grep -E -i "iso|shutter"

KeyError: 'network_g'

python scripts/image_process.py -p 'pretrained/LED_Pretrain_None_None_CVPR20_Setting_Ratio100-300.pth' --data_path '/media/ehooph/data/Dataset/speednoise/raw' --save_path '/media/ehooph/data/Dataset/speednoise/change' -opt 'options/LED/pretrain/CVPR20_ELD_Setting.yaml' --led

我使用这行命令运行时报错
Building network...
Traceback (most recent call last):
File "scripts/image_process.py", line 115, in
image_process()
File "/home/ehooph/anaconda3/envs/LED-ICCV23/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "scripts/image_process.py", line 89, in image_process
network_g = build_network(yaml_load(args.network_options)['network_g'])
KeyError: 'network_g'
请问怎么解决

请教两个问题

1.请问noise_generator.py中的CalibratedNoisyPairGenerator类的forward的参数scale是什么含义呀，比特深度吗
2.您觉得这个噪声模型直接往RGB上加效果如何，或者说这个噪声模型对raw域挺奏效，那对RGB域是否也奏效呢（如果你们还没试的话可以我试了之后可以再交流^_^）

Question about the terminal output

Hi, what a nice work!

I'm going to train other models using this code, but I find there are some warnings in the terminal output. Would you mind helping to check whether such output is the normal output?

Training command (using all the default options except the num_worker_per_gpu and batch_size_per_gpu):

CUDA_VISIBLE_DEVICES=0 python led/train.py -opt options/LED/pretrain/CVPR20_ELD_Setting_Ratio1-200.yaml

Output:

2023-11-28 07:13:14,138 WARNING: Params base_module.conv1_1.weight will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv1_1.bias will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv1_2.weight will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv1_2.bias will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv2_1.weight will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv2_1.bias will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv2_2.weight will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv2_2.bias will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv3_1.weight will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv3_1.bias will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv3_2.weight will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv3_2.bias will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv4_1.weight will not be optimized.
2023-11-28 07:13:14,138 WARNING: Params base_module.conv4_1.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv4_2.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv4_2.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv5_1.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv5_1.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv5_2.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv5_2.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.upv6.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.upv6.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv6_1.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv6_1.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv6_2.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv6_2.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.upv7.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.upv7.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv7_1.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv7_1.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv7_2.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv7_2.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.upv8.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.upv8.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv8_1.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv8_1.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv8_2.weight will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.conv8_2.bias will not be optimized.
2023-11-28 07:13:14,139 WARNING: Params base_module.upv9.weight will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params base_module.upv9.bias will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params base_module.conv9_1.weight will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params base_module.conv9_1.bias will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params base_module.conv9_2.weight will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params base_module.conv9_2.bias will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params base_module.conv10_1.weight will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params base_module.conv10_1.bias will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv1_1.align_weights.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv1_1.align_biases.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv1_2.align_weights.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv1_2.align_biases.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv2_1.align_weights.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv2_1.align_biases.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv2_2.align_weights.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv2_2.align_biases.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv3_1.align_weights.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv3_1.align_biases.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv3_2.align_weights.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv3_2.align_biases.5 will not be optimized.
2023-11-28 07:13:14,140 WARNING: Params repnr_module.conv4_1.align_weights.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv4_1.align_biases.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv4_2.align_weights.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv4_2.align_biases.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv5_1.align_weights.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv5_1.align_biases.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv5_2.align_weights.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv5_2.align_biases.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv6_1.align_weights.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv6_1.align_biases.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv6_2.align_weights.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv6_2.align_biases.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv7_1.align_weights.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv7_1.align_biases.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv7_2.align_weights.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv7_2.align_biases.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv8_1.align_weights.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv8_1.align_biases.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv8_2.align_weights.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv8_2.align_biases.5 will not be optimized.
2023-11-28 07:13:14,141 WARNING: Params repnr_module.conv9_1.align_weights.5 will not be optimized.
2023-11-28 07:13:14,142 WARNING: Params repnr_module.conv9_1.align_biases.5 will not be optimized.
2023-11-28 07:13:14,142 WARNING: Params repnr_module.conv9_2.align_weights.5 will not be optimized.
2023-11-28 07:13:14,142 WARNING: Params repnr_module.conv9_2.align_biases.5 will not be optimized.

calib_cn.md 标定采样问题

我使用您calib_cn.md 文档中的操作流程，我色块采样的时候是用了同样大小的框在图像中局部采样如下所示，但在计算K与var的时候发现，在iso较低时（100-400）k与var并不成线性关系，请问这是正常的现象吗？那如果不成线性关系我该如何继续后续的步骤？还是说只要用一条直线拟合大部分的点就可？不一定需要他们呈线性关系？

关于论文中Table1的数据疑惑

您好，我想问一个问题，关于您论文ELD和SID方法的一些测试数据，好像和ELD论文里及PMN论文里的测试数据对不上？还是说您测试的方法不一致？
比如ELD期刊版本的论文里，paired data训练的结果在ELD的SonyA7S2子集上是大概42.45（x200），ELD建模方法是43.42（x200）。但是我看您的论文里的数据（Table 1）ELD是42.544（x200）。以及在SID的Sony子集上数据好像也对不上。

NAFNet版本LED疑问

您好，我在您提供的的NAFnet预训练网络权重下，进行微调。但是部署的时候图片出现明显的红色噪点，请问您在进行NAFNet部分实验是如何操作的，还是我需要调整finetune部分的细节

关于代码

/scripts/data_preparation/accelerate.sh的第10行，似乎缺少 \ 。

python scripts/data_preparation/eld_to_sid_structure.py \
    --data-path datasets/ICCV23-LED/ELD_sym
    --save-path datasets/ICCV23-LED/ELD

Can I directly infer using my noisy image？

My image isn't from a specific camera; it's a noisy image from the internet

关于SonyA7S2的finetune训练问题

您好，我发现您在SonyA7S2的finetune训练时使用了6对图片，但是前两对图片(x100)的iso都是250，请问是什么原因？
./short/00015_00_0.1s.ARW ./long/00015_00_10s.ARW
./short/00063_00_0.1s.ARW ./long/00063_00_10s.ARW
您提到在进行finetune时选择的2对图片应当在相同的ratio但不同的iso上。

customized rawpy directory

Hi, thanks for your VERY impressive work! I have a problem when I installed customized rawpy of ELD. As said in ELD _" type pip install -e . in the customized rawpy directory", while I can not find the customized rawpy directory provided by ELD. Could you provide some suggestions?

噪声标定方法和噪声参数计算方法

感谢您的分享，针对噪声标定方法和噪声参数计算方法，我也参考了ELD论文中的方法和博客，特别是噪声参数计算方法细节没有讲，大致计算了噪声参数，拟合了K，但是K的范围很小感觉不对。
请问可以参考什么文章有系统的噪声标定方法，以及如何计算获得噪声参数的，来标定自己相机？
谢谢。

关于SID_Sony测试集的问题

作者您好，请问您的论中中给出的在SID_Sony数据集上的测试结果(如下图)是用了SID中所有的不同ratio的测试数据吗？
但您的valid_100.txt,valid_250.txt,valid_300.txt中的数据都不是很多，我不太理解，期待您的回复！

Implementation Details

Hello! Really impressive work! Already starred!
I have 3 questions regarding implementation details.

1. For a UNet structure, do you replace all the convolution blocks with RepNR blocks?
2. Is the CSA^k representing the noise generator, or does the input F_k already transformed by the noise? Quoting from the paper, "CSA focuses on adjusting the distribution of input features." What purpose does CSA serve?
3. Given that the log of system gain is linear with log of the variance of noise, how do we get the noise parameters with 3 paired data? Could you please give me some intuition, or something to read about?
   Thanks again for your work!

Teaser Images

遇到一个缺库问题

您好，我在执行Quick Demo的 python scripts/image_process.py --help 遇到如下的缺库问题， bash install.sh 是正常结束的。请问这个具体是缺什么，有办法解决吗？

(LED-ICCV23) xiexie@xiexie-MSI:~/Project/ICCV23-LED$ python scripts/image_process.py --help
Traceback (most recent call last):
File "scripts/image_process.py", line 4, in
import rawpy
File "/home/xiexie/Project/ICCV23-LED/downloads/rawpy/rawpy/init.py", line 5, in
import rawpy._rawpy
ImportError: libraw.so.23: cannot open shared object file: No such file or directory

关于color bias的获取

想请教一下关于噪声模型中color bias是怎么获取的呢？我看您提供的标定代码中并无color bias的获取逻辑

rawpy bug fix need upgrade

The old code in rawpy/_rawpy.pyx

ndarr.base = <PyObject*> self

through error Assignment to a read-only property

should be changed to issue

np.PyArray_SetBaseObject(ndarr, self)

please checkout the official rawpy implementation.

https://github.com/letmaik/rawpy/blob/main/rawpy/_rawpy.pyx

采用repnr_restormer在finetune时报错

采用repnr_restormer在finetune时报错，log如下：

INFO: End of training. Time consumed: 0:05:02
INFO: Save the latest model.
Traceback (most recent call last):
File "led/train.py", line 223, in
train_pipeline(root_path)
File "led/train.py", line 212, in train_pipeline
model.save(epoch=-1, current_iter=-1) # -1 stands for the latest
File "raw_image_denoising_model.py", line 367, in save
self.save_network([self.net_g, self.net_g.deploy()], 'net_g', current_iter, param_key=['params', 'params_deploy'])
File "anaconda3/envs/pytorch-1.13/lib/python3.7/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "led/archs/repnr_utils.py", line 346, in deploy
deploy(self.base_module, self.repnr_module)
File "led/archs/repnr_utils.py", line 343, in deploy
deploy(mm, getattr(m, n))
File "led/archs/repnr_utils.py", line 343, in deploy
deploy(mm, getattr(m, n))
File "led/archs/repnr_utils.py", line 336, in deploy
m_bak.bias.data = bias
AttributeError: 'NoneType' object has no attribute 'data'

great work!

关于fine-tune的数据ISO要求

你好，你的论文对我非常有帮助，我现在想用自己采集的数据做fine-tune，但是我不太理解关于需要ISO差距较大的数据对。
是场景和场景之间需要不同的ISO，还是LQ和GT之间需要不同的ISO呢？
固定ISO可以做吗？

RuntimeError: Error(s) in loading state_dict for UNetArch:

(led) root@00ee9d88811f:~/LED# python scripts/image_process.py -p checkpoints/pretrained/LED_Pretrain_None_None_CVPR20_Setting_Ratio1-200.pth --data_path images/test/ -opt options/base/network_g/unet.yaml
Building network...
Loading checkpoint...
Loading UNetArch model from checkpoints/pretrained/LED_Pretrain_None_None_CVPR20_Setting_Ratio1-200.pth, with param key: [params].
Traceback (most recent call last):
File "scripts/image_process.py", line 115, in
image_process()
File "/root/anaconda3/envs/led/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "scripts/image_process.py", line 91, in image_process
load_network(network_g, args.pretrained_network, param_key='params' if not args.led else 'params_deploy')
File "scripts/image_process.py", line 37, in load_network
net.load_state_dict(load_net, strict=strict)
File "/root/anaconda3/envs/led/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1671, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for UNetArch:
Missing key(s) in state_dict: "conv1_1.weight", "conv1_1.bias", "conv1_2.weight", "conv1_2.bias", "conv2_1.weight", "conv2_1.bias", "conv2_2.weight", "conv2_2.bias", "conv3_1.weight", "conv3_1.bias", "conv3_2.weight", "conv3_2.bias", "conv4_1.weight", "conv4_1.bias", "conv4_2.weight", "conv4_2.bias", "conv5_1.weight", "conv5_1.bias", "conv5_2.weight", "conv5_2.bias", "conv6_1.weight", "conv6_1.bias", "conv6_2.weight", "conv6_2.bias", "conv7_1.weight", "conv7_1.bias", "conv7_2.weight", "conv7_2.bias", "conv8_1.weight", "conv8_1.bias", "conv8_2.weight", "conv8_2.bias", "conv9_1.weight", "conv9_1.bias", "conv9_2.weight", "conv9_2.bias".
Unexpected key(s) in state_dict: "conv1_1.main_weight", "conv1_1.main_bias", "conv1_1.align_weights.0", "conv1_1.align_weights.1", "conv1_1.align_weights.2", "conv1_1.align_weights.3", "conv1_1.align_weights.4", "conv1_1.align_weights.5", "conv1_1.align_biases.0", "conv1_1.align_biases.1", "conv1_1.align_biases.2", "conv1_1.align_biases.3", "conv1_1.align_biases.4", "conv1_1.align_biases.5", "conv1_2.main_weight", "conv1_2.main_bias", "conv1_2.align_weights.0", "conv1_2.align_weights.1", "conv1_2.align_weights.2", "conv1_2.align_weights.3", "conv1_2.align_weights.4", "conv1_2.align_weights.5", "conv1_2.align_biases.0", "conv1_2.align_biases.1", "conv1_2.align_biases.2", "conv1_2.align_biases.3", "conv1_2.align_biases.4", "conv1_2.align_biases.5", "conv2_1.main_weight", "conv2_1.main_bias", "conv2_1.align_weights.0", "conv2_1.align_weights.1", "conv2_1.align_weights.2", "conv2_1.align_weights.3", "conv2_1.align_weights.4", "conv2_1.align_weights.5", "conv2_1.align_biases.0", "conv2_1.align_biases.1", "conv2_1.align_biases.2", "conv2_1.align_biases.3", "conv2_1.align_biases.4", "conv2_1.align_biases.5", "conv2_2.main_weight", "conv2_2.main_bias", "conv2_2.align_weights.0", "conv2_2.align_weights.1", "conv2_2.align_weights.2", "conv2_2.align_weights.3", "conv2_2.align_weights.4", "conv2_2.align_weights.5", "conv2_2.align_biases.0", "conv2_2.align_biases.1", "conv2_2.align_biases.2", "conv2_2.align_biases.3", "conv2_2.align_biases.4", "conv2_2.align_biases.5", "conv3_1.main_weight", "conv3_1.main_bias", "conv3_1.align_weights.0", "conv3_1.align_weights.1", "conv3_1.align_weights.2", "conv3_1.align_weights.3", "conv3_1.align_weights.4", "conv3_1.align_weights.5", "conv3_1.align_biases.0", "conv3_1.align_biases.1", "conv3_1.align_biases.2", "conv3_1.align_biases.3", "conv3_1.align_biases.4", "conv3_1.align_biases.5", "conv3_2.main_weight", "conv3_2.main_bias", "conv3_2.align_weights.0", "conv3_2.align_weights.1", "conv3_2.align_weights.2", "conv3_2.align_weights.3", "conv3_2.align_weights.4", "conv3_2.align_weights.5", "conv3_2.align_biases.0", "conv3_2.align_biases.1", "conv3_2.align_biases.2", "conv3_2.align_biases.3", "conv3_2.align_biases.4", "conv3_2.align_biases.5", "conv4_1.main_weight", "conv4_1.main_bias", "conv4_1.align_weights.0", "conv4_1.align_weights.1", "conv4_1.align_weights.2", "conv4_1.align_weights.3", "conv4_1.align_weights.4", "conv4_1.align_weights.5", "conv4_1.align_biases.0", "conv4_1.align_biases.1", "conv4_1.align_biases.2", "conv4_1.align_biases.3", "conv4_1.align_biases.4", "conv4_1.align_biases.5", "conv4_2.main_weight", "conv4_2.main_bias", "conv4_2.align_weights.0", "conv4_2.align_weights.1", "conv4_2.align_weights.2", "conv4_2.align_weights.3", "conv4_2.align_weights.4", "conv4_2.align_weights.5", "conv4_2.align_biases.0", "conv4_2.align_biases.1", "conv4_2.align_biases.2", "conv4_2.align_biases.3", "conv4_2.align_biases.4", "conv4_2.align_biases.5", "conv5_1.main_weight", "conv5_1.main_bias", "conv5_1.align_weights.0", "conv5_1.align_weights.1", "conv5_1.align_weights.2", "conv5_1.align_weights.3", "conv5_1.align_weights.4", "conv5_1.align_weights.5", "conv5_1.align_biases.0", "conv5_1.align_biases.1", "conv5_1.align_biases.2", "conv5_1.align_biases.3", "conv5_1.align_biases.4", "conv5_1.align_biases.5", "conv5_2.main_weight", "conv5_2.main_bias", "conv5_2.align_weights.0", "conv5_2.align_weights.1", "conv5_2.align_weights.2", "conv5_2.align_weights.3", "conv5_2.align_weights.4", "conv5_2.align_weights.5", "conv5_2.align_biases.0", "conv5_2.align_biases.1", "conv5_2.align_biases.2", "conv5_2.align_biases.3", "conv5_2.align_biases.4", "conv5_2.align_biases.5", "conv6_1.main_weight", "conv6_1.main_bias", "conv6_1.align_weights.0", "conv6_1.align_weights.1", "conv6_1.align_weights.2", "conv6_1.align_weights.3", "conv6_1.align_weights.4", "conv6_1.align_weights.5", "conv6_1.align_biases.0", "conv6_1.align_biases.1", "conv6_1.align_biases.2", "conv6_1.align_biases.3", "conv6_1.align_biases.4", "conv6_1.align_biases.5", "conv6_2.main_weight", "conv6_2.main_bias", "conv6_2.align_weights.0", "conv6_2.align_weights.1", "conv6_2.align_weights.2", "conv6_2.align_weights.3", "conv6_2.align_weights.4", "conv6_2.align_weights.5", "conv6_2.align_biases.0", "conv6_2.align_biases.1", "conv6_2.align_biases.2", "conv6_2.align_biases.3", "conv6_2.align_biases.4", "conv6_2.align_biases.5", "conv7_1.main_weight", "conv7_1.main_bias", "conv7_1.align_weights.0", "conv7_1.align_weights.1", "conv7_1.align_weights.2", "conv7_1.align_weights.3", "conv7_1.align_weights.4", "conv7_1.align_weights.5", "conv7_1.align_biases.0", "conv7_1.align_biases.1", "conv7_1.align_biases.2", "conv7_1.align_biases.3", "conv7_1.align_biases.4", "conv7_1.align_biases.5", "conv7_2.main_weight", "conv7_2.main_bias", "conv7_2.align_weights.0", "conv7_2.align_weights.1", "conv7_2.align_weights.2", "conv7_2.align_weights.3", "conv7_2.align_weights.4", "conv7_2.align_weights.5", "conv7_2.align_biases.0", "conv7_2.align_biases.1", "conv7_2.align_biases.2", "conv7_2.align_biases.3", "conv7_2.align_biases.4", "conv7_2.align_biases.5", "conv8_1.main_weight", "conv8_1.main_bias", "conv8_1.align_weights.0", "conv8_1.align_weights.1", "conv8_1.align_weights.2", "conv8_1.align_weights.3", "conv8_1.align_weights.4", "conv8_1.align_weights.5", "conv8_1.align_biases.0", "conv8_1.align_biases.1", "conv8_1.align_biases.2", "conv8_1.align_biases.3", "conv8_1.align_biases.4", "conv8_1.align_biases.5", "conv8_2.main_weight", "conv8_2.main_bias", "conv8_2.align_weights.0", "conv8_2.align_weights.1", "conv8_2.align_weights.2", "conv8_2.align_weights.3", "conv8_2.align_weights.4", "conv8_2.align_weights.5", "conv8_2.align_biases.0", "conv8_2.align_biases.1", "conv8_2.align_biases.2", "conv8_2.align_biases.3", "conv8_2.align_biases.4", "conv8_2.align_biases.5", "conv9_1.main_weight", "conv9_1.main_bias", "conv9_1.align_weights.0", "conv9_1.align_weights.1", "conv9_1.align_weights.2", "conv9_1.align_weights.3", "conv9_1.align_weights.4", "conv9_1.align_weights.5", "conv9_1.align_biases.0", "conv9_1.align_biases.1", "conv9_1.align_biases.2", "conv9_1.align_biases.3", "conv9_1.align_biases.4", "conv9_1.align_biases.5", "conv9_2.main_weight", "conv9_2.main_bias", "conv9_2.align_weights.0", "conv9_2.align_weights.1", "conv9_2.align_weights.2", "conv9_2.align_weights.3", "conv9_2.align_weights.4", "conv9_2.align_weights.5", "conv9_2.align_biases.0", "conv9_2.align_biases.1", "conv9_2.align_biases.2", "conv9_2.align_biases.3", "conv9_2.align_biases.4", "conv9_2.align_biases.5".
模型权重无法加载