search

ynu-yangpeng / GLMC

[CVPR2023] Global and Local Mixture Consistency Cumulative Learning for Long-tailed Visual Recognitions
69 stars 12 forks source link

training details on ImageNet-LT #9

Open WilyZhao8 opened 12 months ago

WilyZhao8 commented 12 months ago

I used the recommended training command (python main.py --dataset ImageNet-LT -a resnext50_32x4d --num_classes 1000 --beta 0.5 --lr 0.1 --epochs 135 -b 120 --momentum 0.9 --weight_decay 2e-4 --resample_weighting 0.2 --label_weighting 1.0 --contrast_weight 10) and the results obtained (Prec@1 54.946 Prec@5 77.946) differ significantly from those published in the paper. May I ask what is the command for training ImageNet?

WakingHours-GitHub commented 3 months ago

I have noticed that the obtained results (Prec@1 55.306, Prec@5 78.6989) are lower than the original paper's results (56.3) by 1%.