How to train my own model with ms-issim? just change the lambda? thanks for answer my question !!

[Jerrry-Li]

Feature

Motivation

Additional context

[YodaEmbedding]

First train using MSE, then fine-tune with MS-SSIM.

https://interdigitalinc.github.io/CompressAI/zoo.html

MS-SSIM optimized networks were fine-tuned from pre-trained MSE networks (with a learning rate of 1e-5 for both optimizers).

[Jerrry-Li]

First train using MSE, then fine-tune with MS-SSIM.

https://interdigitalinc.github.io/CompressAI/zoo.html

MS-SSIM optimized networks were fine-tuned from pre-trained MSE networks (with a learning rate of 1e-5 for both optimizers).

Thanks you !!! But i still don't know how to use it. Can you show me your code? just for ms-ssim training .When I training my own model ,how to use the ms-ssim metric

[Do1e]

@Freed-Wu have committed a PR which includes ms-ssim loss. You can check his code and use the RateDistortionLoss.

[fracape]

@Freed-Wu have committed a PR which includes ms-ssim loss. You can check his code and use the RateDistortionLoss.

Yes, I merged it. Note that I haven't added the option in examples/codec.py but it's straightforward to add

[Jerrry-Li]

One more question is, where did you download the CLIC2020-PRO dataset you used and can you provide a link? I have read many articles where the BPG data on CLIC2020-PRO is different from the data you provided