#ResNeXt50_vd_64x4d python train.py \ --model=ResNeXt50_vd_64x4d \ --batch_size=256 \ --lr_strategy=cosine_decay \ --lr=0.1 \ --num_epochs=200 \ --model_save_dir=output/ \ --l2_decay=1e-4 \ --use_mixup=True \ --use_label_smoothing=True \ --label_smoothing_epsilon=0.1 \
