/Minimum-Probability-Penalty

Primary LanguagePython

Minimum-Probability-Penalty

Usage

  1. Download CUB-200, FGVC-Aircrafts and Stanford Cars.
  2. Convert the dataset to the designated format using utils/convert_data.py and place them in data/bird(aircraft/car)
  3. Run the following code to get the reported results:

CUB

python train.py --model-name=densenet161 --scheduler=cos --lr=0.008 --loss=mpp --smoothing=0.1 -b=16 --image-size=600 --weight-decay=5e-4

Aircrafts

python train.py --model-name=densenet161 --lr=0.005 --loss=mpp --smoothing=0.1 -b=4 --weight-decay=5e-4

Stanford Cars

python train.py --model-name=densenet161 --lr=0.001 --loss=mpp --smoothing=0.1 -b=4 --weight-decay=5e-4