5663015 / Imagenet_training

Use several models to train Imagenet

Imagenet_training

训练

export CUDA_VISIBLE_DEVICES=6,7 && nohup python -u train.py \
	--arch model_name \
	--data_backend pytorch \
	--epochs 150 \
	--batch_size 256 \
	--lr 0.05 \
	--weight_decay 4e-5 \
	--dropout 0.2 \
	--drop_connect 0.2 \
	--checkpoint your_path > xx.log 2>&1 &

实验模型与结果

模型	accuracy (top-1) / %	running time (2 V100 GPUs)
EfficientNet_b0	74.1	about 66 hours
MixNet_m	73.15	about 92 hours
MobileNet_v3	72.042	about 52 hours
MNASNet_a1	72.342	about 51 hours