hysts / pytorch_mpiigaze

An unofficial PyTorch implementation of MPIIGaze and MPIIFaceGaze
MIT License
346 stars 85 forks source link

results of run_all_mpiigaze_resnet_preact.sh #30

Closed yuyaya-foifoi closed 3 years ago

yuyaya-foifoi commented 3 years ago

Thank you for sharing your great work.

I have a question about the evaluation results of run_all_mpiigaze_resnet_preact.sh. That is, why is the angle error low something like in the range of 1 to 2?

Based on a code, My understanding is that run_all_mpiigaze_resnet_preact.sh evaluates test_id's data with a model trained on data that excludes test_id's data.

So this is not something similar to your previous comment, I guess.

Thank you

hysts commented 3 years ago

Hi, @yuyaya-foifoi

Hmm... That's strange. Something must be wrong, but I have no idea what.

yuyaya-foifoi commented 3 years ago

Thanks for the reply. Just to confirm, the result of MPIIFaceGaze on README was calculated by running run_all_mpiigaze_resnet_preact.sh as I did?

hysts commented 3 years ago

@yuyaya-foifoi

If you ran the code without any modification, that's the case. I ran the code again just in case, and got the similar results in the README. Are you sure you are looking at the test results, not the val?

yuyaya-foifoi commented 3 years ago

Thank you

Quick check,

right?

hysts commented 3 years ago

Yes.

yuyaya-foifoi commented 3 years ago

Then I'm checking the right txt file.

I'll retry

Download the dataset and preprocess it

from this process again and see what happens.

yuyaya-foifoi commented 3 years ago

Hmm, somehow I could get a result something similar to the result written in README. Bit weird but my problem was solved so I'll close this issue.

スクリーンショット 2021-03-12 14 05 55