After running the program, the performance of the model decreases.

Simon4Yan / eSPGAN

We extend the SPGAN (our prior work in CVPR2018) to an end-to-end version, named eSPGAN. https://arxiv.org/pdf/1811.10551.pdf

MIT License

47 stars 12 forks source link

After running the program, the performance of the model decreases. #10

Open yxsysu opened 3 years ago

yxsysu commented 3 years ago

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!

yxsysu commented 3 years ago

I read the paper, and I found that the problem may come from here: 'Specifically, we normalize the image with the same mean (0.5, 0.5, 0.5) and standard deviation (0.5, 0.5, 0.5) for both the image translator and the feature learner' while the mean and std used for normalization in 'https://github.com/Simon4Yan/Person_reID_baseline_pytorch' is [0.485, 0.456, 0.406], [0.229, 0.224, 0.225].