airsplay / lxmert

PyTorch code for EMNLP 2019 paper "LXMERT: Learning Cross-Modality Encoder Representations from Transformers".
MIT License
923 stars 157 forks source link

np.all(feature == 0) is True, why #103

Open Yinhance opened 3 years ago

Yinhance commented 3 years ago

I use the command CUDA_VISIBLE_DEVICES=0 python extract_nlvr2_image.py --split test but when I output the features: pool5[keep_boxes],I found these are equal to 0,like: [0., 0., 0., ..., 0., 0., 0.], [0., 0., 0., ..., 0., 0., 0.], [0., 0., 0., ..., 0., 0., 0.], [0., 0., 0., ..., 0., 0., 0.]

What's the reason for this problem?Thx a lot!