yelboudouri / EmoNeXt

Code for the paper: "EmoNeXt: an Adapted ConvNeXt for Facial Emotion Recognition"
28 stars 8 forks source link

Cannot reproduce the accuracy in the paper #6

Open chn-lee-yumi opened 4 months ago

chn-lee-yumi commented 4 months ago

image

I only got 71.4127% in the Tiny model.

Here's my command:

python3 train.py      --dataset-path='FER2013'      --batch-size=64 --lr=0.0001      --epochs=300      --amp      --in_22k      --num-workers=6      --model-size='tiny'

I also modified scheduler.py, changed max_lr from 1e-3 to 1e-4.

1351986826 commented 3 weeks ago

I also only got an accuracy rate of 71.6077, may I ask if you have improved since then

chn-lee-yumi commented 3 weeks ago

I also only got an accuracy rate of 71.6077, may I ask if you have improved since then

Unfortunately, I am stuck here. I don't know how to improve too.