search

IIGROUP / MANIQA

[CVPRW 2022] MANIQA: Multi-dimension Attention Network for No-Reference Image Quality Assessment
Apache License 2.0
307 stars 36 forks source link

Regarding loss convergence? #31

Closed Mishra1995 closed 1 year ago

Mishra1995 commented 1 year ago

Hello authors,

Thanks for open-sourcing your code and models! It really helps in reproducing the results.

My query is regarding the loss convergence. I tried to train the model on KonIQ dataset before you modified the code and it seems the network is overfitting a lot with a huge gap between training and validation loss. I even tried to change the loss function and did normalisation but it didn't factor.

Any suggestions regarding the same? Can you also share the loss plots to verify the convergence criteria?

Thanks!

TianheWu commented 1 year ago

When we train the MANIQA on Koniq10k dataset, we resize the image into 224x224. The model will reach the score in our checkpoints report. The loss value will converge.

Mishra1995 commented 1 year ago

MANIQA_origina

This is the loss plot I am getting after running the current code on KonIQ dataset for 63 epochs without changing any parameters. Is there anything else that needs to be done?

TianheWu commented 1 year ago

You can test your model's ability in your test set.