ChengyueGongR / PatchVisionTransformer

71 stars 13 forks source link

Do you need to train 600 epoch totally? #2

Open wangpichao opened 3 years ago

wangpichao commented 3 years ago

According to your instruction in readme, do you need to train 600 epoch totally to get the results shown in the paper?

ChengyueGongR commented 3 years ago

Hi, As listed in the paper, overall 400 epochs. 300 epochs with a drop path rate equal to 0.5, and additional 100 epochs with drop path rate equal to 0.75, --start_epoch 300 --epochs 400.

We currently work on some more strong models and will release the full log file, checkpoint and train commands for these strong models soon.

GoJunHyeong commented 1 year ago

Hi, can you share the checkpoint and log files for DeiT-Small-24?