simon4yan / espgan Goto Github PK

您好！很荣幸拜读您的论文，我现在有一些关于espgan和spgan区别的疑惑，看了代码后，我发现您在espgan.py里没有使用孪生网络，当然也没有使用论文中的对比损失函数来训练它，直接用一个训练好的基准网络代替，那么espgan不就无法实现spgan 中的相似度保留，不知我理解的是否有偏差，也希望您能告诉我他们俩之间的根本区别，期待您的回复！

Hi, any plan to release the code for eSPGAN?

Hi,
could you plz tell me how many data we need to train eSPGAN?
thanks

Dear authors, When will the code be released？

Thanks for sharing the code. How do you test the re-ID model? I found that the IDE adds a classifier directly after the output of resnet. Is this classifier removed directly for testing?

Hi, thanks for your work, I find the parameter is different with the tensorflow, such as the learning rate ,tne batchsize and the crop_width

the training and testing codes of spgan exist two or more bugs.

1. https://github.com/Simon4Yan/eSPGAN/blob/master/py-spgan/train_spgan.py#L104
2. https://github.com/Simon4Yan/eSPGAN/blob/master/py-spgan/test_spgan.py#L58

root@f2aaf8a01a08:/home/SPGAN-torch/py-spgan# CUDA_VISIBLE_DEVICES=0 python train_spgan.py --dataset=market2duke --gpu_id=0
Traceback (most recent call last):
File "train_spgan.py", line 6, in
import models.models_spgan as models
File "/home/SPGAN-torch/py-spgan/models/models_spgan.py", line 11, in
import models.model_imagenet
ModuleNotFoundError: No module named 'models.model_imagenet'
是否是少了一个包叫：model_imagenet
希望能增加readme文件，比如一些包的版本信息数据，存放的格式等等
非常感谢！

Hi Weijian,
Thanks for your released code and your SPGAN work.
I got confused about the data preparation process. Is the train_A and test_A(train_B and test_B) are the same data in SPGAN? As to the person re-ID dataset, the style transfer is performed during training and training a discriminate model on both re-ID datasets and style transferred datasets.
Thank for your kindly reply.

I ran the code from 'https://github.com/Simon4Yan/Person_reID_baseline_pytorch' using training data on Duke and obtain a model with 48.8% rank1 accuracy on Market. But after running espgan.py with the default parameters, the model only achieved 41.0 rank1 accuracy on Market...
I want to know what may have gone wrong. Thanks a lot!