search

kdhht2334 / AVCE_FER

[ECCV2022] The official repository of Emotion-aware Multi-view Contrastive Learning for Facial Emotion Recognition
23 stars 1 forks source link

the dataset on which the model "Sparse_..._.t7" in AVCE_demo is trained? #4

Closed MagicGome closed 1 year ago

MagicGome commented 1 year ago

May I inquire about the dataset on which the model "Sparse....t7" in AVCE_demo is trained? I have noticed a significant disparity between the predicted outcomes and the labels when evaluating it on datasets like AFEW-VA. Could you kindly provide some insights into this matter?

kdhht2334 commented 1 year ago

Hi,

Weights in demo folder are trained using both AFEW-VA and Aff-wild.

That's why significant disparity may happen.

Please also note that you have to do preprocessing procedure (e.g. face crop and remove fade in/out scene).

MagicGome commented 1 year ago

Could you please provide your preprocessing code? It seems that there isn't any method mentioned in https://pytorch.org/tutorials/beginner/data_loading_tutorial.html for face crop or removing fade in/out scenes. Thank you very much for your assistance and time.

kdhht2334 commented 1 year ago

Here you are.

In my case, all data are preprocessed by MTCNN.

Please see https://github.com/kdhht2334/AVCE_FER/blob/main/preprocessing.py

MagicGome commented 1 year ago

Thanks a lot