MLBNoAtt does not use BayesianGRU

[backpropper]

In the default case (I guess MLBNoAtt), it uses normal GRU and not the Bayesian one. The README says otherwise.

[Cadene]

You are right, the default.yaml uses a GRU, not a Bayesian one:

• https://github.com/Cadene/vqa.pytorch/blob/master/options/vqa2/default.yaml#L23
• https://github.com/Cadene/vqa.pytorch/blob/master/options/vqa/default.yaml#L22

Here, a BayesianGRU is used:

• https://github.com/Cadene/vqa.pytorch/blob/master/options/vqa/mutan_noatt_train.yaml#L22

I am sorry for this misleading information.

[Cadene]

I just added two options:

• https://github.com/Cadene/vqa.pytorch/blob/master/options/vqa/mlb_noatt_train.yaml
• https://github.com/Cadene/vqa.pytorch/blob/master/options/vqa2/mlb_noatt_train.yaml

[backpropper]

Thanks!