princeton-vl/px2graph

predict/valid

Closed this issue · 0 comments

beihangwenbin commented

I train the model with sg_task as SG ,but the results are too bad.
train: python main.py -e second --sg_task=SG --batchsize=20 --gpu_choice='0,1'
predict: python main.py --branch second -e my_results --predict valid
python eval.py my_results/valid_preds
R@50: 0.03502271674069844
R@100: 0.04519653756966068
It' OK?