facebookresearch / keypointnerf Goto Github PK

Hi, author. Thanks for your great work!

I'm trying to use your pre-trained model to render my own data, but the function "decode_batch" and the parameters "batch" seems too complicate for me. I'm wondering if there is some easy way to do it?

Look forwarding to your reply!

Hi, thanks for your nice work. I download zju_mocap dataset but I dont find CoreView_396 as mentioned in your dataset.

we can not load environment.yml,
ResolvePackageNotFound:

• libffi==3.4.2=h7f98852_5
• setuptools==59.4.0=py38h578d9bd_0
• libzlib==1.2.11=h36c2ea0_1013
• intel-openmp==2021.4.0=h06a4308_3561
• zlib==1.2.11=h36c2ea0_1013
• mkl_fft==1.3.1=py38h8666266_1
• zstd==1.5.0=ha95c52a_0
  ...
  can you pack all env data to netdisk, thanks!

Hi,

Thank you for releasing the code.

I wonder how much GPU memory KeypointNeRF requires if I use more images like 16 images. Since keypointnerf needs to keep feature maps of all imput images, I want to know if 24GB GPU memory is enough for the training.

Thanks

Hi, authors

which tool are you used to generate the zju_joints3d.zip?

I am trying to train KeypointNeRF on my own dataset.

Hi, author. Thanx for your great work!
It seems that the example given is on task of reconstruction of human bodies. I wonder how to reconstruct human heads from two or three images.
Maybe I read it wrong. Look forwarding to your reply!

Is there a dataset for 3D facial reconstruction? How to generate 3d face obj files? thank you！！！

Thanks for your well-organized code. When I run train.py, an error is as follow:

Could you tell me how to solve it?

hi @markomih ,

Thank you for releasing the code.

Can you share the hardware configuration of your training?

i used a NVIDIA3090 card to run traning code, the errors are as follows:

$ nvidia-smi 
Thu Jun 29 16:09:01 2023       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.105.17   Driver Version: 525.105.17   CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  Off  | 00000000:B1:00.0 Off |                  Off |
|  0%   41C    P0   103W / 450W |      0MiB / 24564MiB |      2%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+


$ python train.py --config ./configs/zju.json --data_root ./data/zju_mocap

RuntimeError: CUDA out of memory. Tried to allocate 222.00 MiB (GPU 0; 23.68 GiB total capacity; 20.55 GiB already allocated; 49.00 MiB free; 21.58 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Hi, authors,

Thanks for your great works! I'm interesting with the performance of KeypointNeRF on the ZJU dataset. I follow the DATA_PREP to prepare dataset and start to train KeypointNeRF.

However, I had the following error:

This is caused when I built the val or test dataset. I had checked the code and found the reason:

I notice that CoreView313 and CoreView315 use different sampled cameras with other persons, see here. The number of cameras in the CoreView 313 and CoreView 315 are 21, so the index of 21 at here is out of bounds.

Could plz to check this for me?

If I have SMPL vertices, how to precalculate 3D joints？

Hi, thanks for the great work!!!

What if the 3D points out of index (let's say it may occur negative indexes of image pixel) after projected to other camera views rather than target views?

Hello, thanks for this great work!

I want to try this method on NeRF synthetic dataset (https://drive.google.com/drive/folders/1JDdLGDruGNXWnM1eqY1FNL9PlStjaKWi).

Is this method applicable to non-human data?

Hey，awesome work！
Can I using the KeypointNERF to reconstruct
a full body which can be exported as a obj file with texture.

I really appreciate the project. Unfortunately, I'm restricted by CUDA memory and limited skills. Could you provide files about the training and inferencing in multiple GPU modes?