chenhao2345 / gcl Goto Github PK
View Code? Open in Web Editor NEWImplementation for CVPR2021 paper "Joint Generative and Contrastive Learning for Unsupervised Person Re-identification"
License: MIT License
Implementation for CVPR2021 paper "Joint Generative and Contrastive Learning for Unsupervised Person Re-identification"
License: MIT License
Couldyou please tell me whether the pretrained identity encoder network you are training in stage 2 is idnet or gen_ pth\dis_.th? For this problem, I can modify the K value without reporting an error, but I feel that it will affect the training accuracy.
Epoch 1 has 10408 labeled samples of 234 ids and 2528 unlabeled samples
Traceback (most recent call last):
File "examples/main.py", line 381, in
main()
File "examples/main.py", line 151, in main
main_worker(args)
File "examples/main.py", line 279, in main_worker
feat_nv2recon, f, f_recon, f_nv, f_nv2recon, pid, index, config, iterations)
File "/home/ly/yx/GCL/gcl/trainer.py", line 112, in gen_update
self.memory_loss_id = self.memory(F.normalize(f), F.normalize(f_nv), index)
File "/home/ly/anaconda3/envs/GCL/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "/home/ly/yx/GCL/gcl/models/memory.py", line 66, in forward
negatives = torch.stack(negatives)
RuntimeError: stack expects each tensor to be equal size, but got [8192] at entry 0 and [7833] at entry 3
你好,现在我想用预训练模型跑generate_data.py生成数据
python examples/generate_data.py
提示我没有outputs/market_init_JVTC_unsupervised/checkpoints没有模型,是否可以提供预训练模型网盘,谢谢
训练模型需要太大的显存,起不来
Hi, I have a few problem with generating synthetic persons using your code in examples/generate_data.py.!
This is the resuls what I get from original image to generated image.
As you can see, the result is not good..
I have resumed your model same as when ./train_stage3_market.sh starts.
Below is the model(generator, discrimminator) weights when ./train_stage2_market.sh ends.
So, I resumed your model from those weights from the code below
"
trainer = DGNet_Trainer(config, model_1, args.idnet_fix).cuda()
iterations = trainer.resume(checkpoint_path, hyperparameters=config)
"
Actually, I trained your model with stage3, -> ./train_stage3_market.sh
and saved the model_best.pth in the outputs folder like below images.
So, I tried to bring your model_best.pth and resume it, hopefully thinking it will generate good results, but I could not resume your model because of the difference of the weights and models,.
What should I do to generate good results of your training datasets??
Thanks in advance!
Epoch 1 has 10408 labeled samples of 234 ids and 2528 unlabeled samples
Traceback (most recent call last):
File "examples/main.py", line 381, in
main()
File "examples/main.py", line 151, in main
main_worker(args)
File "examples/main.py", line 279, in main_worker
feat_nv2recon, f, f_recon, f_nv, f_nv2recon, pid, index, config, iterations)
File "/home/ly/yx/GCL/gcl/trainer.py", line 112, in gen_update
self.memory_loss_id = self.memory(F.normalize(f), F.normalize(f_nv), index)
File "/home/ly/anaconda3/envs/GCL/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "/home/ly/yx/GCL/gcl/models/memory.py", line 66, in forward
negatives = torch.stack(negatives)
RuntimeError: stack expects each tensor to be equal size, but got [8192] at entry 0 and [7833] at entry 3
Hello.
I have some issue with your model with different torch version.
When I trained your network with torch version 1.6.0, the performance was really degraded.
Stage3 first epoch results and losses are as below.
pytorch == 1.2.0 (Your original version)
As you can see, it is so differenent when I only changed torch version.
I am trying to figure out what caused the problem.
Could you give me some help what is causing the difference? I need to inevitably use torch version higher than 1.6.0..
Thanks in advance.
The fid is not described in Fig 2. If it can be decripted, the readability would be improved.
Thank you for sharing your great work.
I have a few questions about the 3D Mesh generator!.
Thanks in advance.
Thanks you for providing your great work.
I read your code and there seems to be no code stage=1, warming up identity encoder.
It saids in the paper that
"We firstly use a state-of-the-art unsupervised Re - ID method to warm up Eid, which is then
considered as a baseline in our ablation studies" which you used JTCV? method to warm up identity encoder."
I figured out when input image x is fed into identity-encoder(Eid), it outputs feature vector fid(2048x4x1 feature after part average pooling in ft_net in the code). But how did you train the encoder(Eid)?? Did you use just apply your ft_net
in the JTCV, or MLC code and train it??
For example, below is the framework of MLC paper.
In the above image, the CNN layer is your ft_net and you trained the whole network so the weights in the CNN(ft_net) could be warmed up?
Is this right??
I have trained and obtained your pretrained model,
how do I run the test result in the paper?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.