search

fengyurenpingsheng / Learned-image-compression-with-GLLMM

The code of the Learned Image Compression with Discretized Gaussian-Laplacian-Logistic Mixture Model and Concatenated Residual Modules
21 stars 5 forks source link

Questions about reproduce #1

Open xiaobaoxia opened 3 years ago

xiaobaoxia commented 3 years ago

Hello, I want to reproduce the results of your MS-SSIM experiment, but the performance is always not good. It may be because there is an error in the loss function. If it is convenient, can you release the code of the training part?

fengyurenpingsheng commented 3 years ago

For MS-SSIM training, you can use the loss function of PSNR to initialize, for example, train tens of thousands of rounds first, and then use the loss function of MS-SSIM for training. There is a library of MS-SSIM in tensorflow. But I am sorry that my current training code is not convenient to open source.

fengyurenpingsheng commented 3 years ago

Hello, I want to reproduce the results of your MS-SSIM experiment, but the performance is always not good. It may be because there is an error in the loss function. If it is convenient, can you release the code of the training part?

For MS-SSIM training, you can use the loss function of PSNR to initialize, for example, train tens of thousands of rounds first, and then use the loss function of MS-SSIM for training. There is a library of MS-SSIM in tensorflow. But I am sorry that my current training code is not convenient to open source.

xiaobaoxia commented 3 years ago

Thank you very much for your reply, it has given me a lot of help. I will try this training strategy. ^_^