Open greathope opened 6 years ago
@greathope As far as I remember, we got those results on the testset (with our old Torch7 code, but you should get the same results with this code in pytorch):
Overall Accuracy is: 64.77
Per Answer Type Accuracy is the following:
other : 54.47
number : 38.43
yes/no : 83.81
@Cadene with trainval datasets? Actually I got 65.42...
That's good news, isn't it?
It might be related to this little change in architecture: https://github.com/Cadene/vqa.pytorch/issues/18
What do you think? @greathope
@Cadene how about: vqav1.0 python train.py --vqa_trainsplit trainval --path_opt options/vqa/mlb_att_trainval.yaml, what accuracy did you acheived?
with this command on vqa v1.0: python train.py --vqa_trainsplit trainval --path_opt options/vqa/mutan_att_trainval.yaml