Car196-ResNet152

Acc 88%

Prepare for the data:

mkdir data
cd data
wget http://imagenet.stanford.edu/internal/car196/car_ims.tgz
wget http://imagenet.stanford.edu/internal/car196/cars_annos.mat

And run:

!python main.py -e 400 -b 36 -lr 0.0001 -m ResNet152 -d 1 --resume 'result/checkpoint.pth.tar'

You can also choose ResNet50 use :

!python main.py -e 400 -b 36 -lr 0.0001 -m ResNet50 -d 1 --resume 'result/checkpoint.pth.tar'

There tells the hard tuning process : Hard param tuning and a little rethink for data augmentation

And I'm sure 88% is not the best acc. For this task is only a benchmark for later work, if time allows I may try to tune it better.