VQA: Limitations in questions and answers

[fizahkhalid]

I want my Finetuned VQA model to be able to answer questions is was not trained on before and similarly provides answers that does not exist in the original answer list (test json file answers in a list).

Is there a limitation to the kind of questions i can the model? If yes, how can I tweak the code to meet my needs?

[zengyan-97]

Hi,

you need to modify the inference process of the VQA model.

do not use this to rank the candidate answers: https://github.com/zengyan-97/X-VLM/blob/master/models/model_vqa.py#L144

instead, you should make it a real generation. for example, you can refer to: https://github.com/zengyan-97/X-VLM/blob/master/models/model_captioning.py#L75

[chilljudaoren]

Traceback (most recent call last):
File "/home/czh/.pycharm_helpers/pydev/pydevd.py", line 1534, in _exec
pydev_imports.execfile(file, globals, locals) # execute the script
File "/home/czh/.pycharm_helpers/pydev/_pydev_imps/_pydev_execfile.py", line 18, in execfile
exec(compile(contents+"\n", file, 'exec'), glob, loc)
File "/home/czh/BLIP-main/Eval_VQA.py", line 261, in
main(args, config)
File "/home/czh/BLIP-main/Eval_VQA.py", line 180, in main
model.load_pretrained(ckpt, config, is_eval=True)
File "/home/czh/BLIP-main/models/XVLM/model_vqa.py", line 89, in load_pretrained
msg = self.load_state_dict(state_dict, strict=False)
File "/home/czh/.conda/envs/czh/lib/python3.9/site-packages/torch/nn/modules/module.py", line 2215, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for XVLM:
size mismatch for vision_encoder.layers.0.blocks.0.attn.relative_position_bias_table: copying a param with shape torch.Size([841, 4]) from checkpoint, the shape in current model is torch.Size([529, 4]).