About dataset

[ProudZhangXiaolong]

How to get novel_vqa_val_known_question.json and novel_vqa_val_known_annotation.json?

[ProudZhangXiaolong]

Thank you very much !!

[szzexpoi]

You can find the data here, as listed in the instruction of the data folder:

https://drive.google.com/file/d/1dTg5Dn1BmCiwY_gXOG06lo60LOZJrw29/view?usp=drivesdk

[ProudZhangXiaolong]

Thank you very much !!

[ProudZhangXiaolong]

Excuse me, where does the npy file come from? Is it processed by trainval_feature.h5?

[szzexpoi]

Yes, please refer to the last several lines of the instruction in the data folder.

[ProudZhangXiaolong]

Thank you!

[ProudZhangXiaolong]

I'm really sorry to bother you again. I want to retrain proto_learning. When I was running extract-vqa. py, np.frombuffer (base64.decodestring)... encountered an error and decoding problem was difficult to solve. Could you share the complete weight file of Prototype_Module to predict labels? (on all datasets). Thank you! My email: zjinhao2022@126.com

[szzexpoi]

You probably need to run that code with python2 (maybe). The complete weights for different datasets can be found in the link under the proto_learning folder: https://drive.google.com/file/d/1alCvI8tub2yv0lJI2shjTlC8fA_TEZLe/view

[ProudZhangXiaolong]

I'm very sorry to bother you again May I ask if it's possible to share the model_best.pt of the README.md under the proto_learning file

import torch proto = torch.load('./ckpt/model_best.pt')['prototype.weight'] torch.save(proto,'prototype.pt')

Thank you very much !

[szzexpoi]

The "model_best.pt" in the instruction is just an example, you should replace it with the weights for different datasets provided in Google drive. Sorry for the confusion.

[ProudZhangXiaolong]

I'm sorry to bother you again, perhaps you misunderstood my meaning. The shape of the parameter of prototype_vqa.pt you provided is [1000,2048], which corresponds to the parameter of the self.prototype in the Prototype Module class. However, I need the complete parameters of the Prototype Module class including self.prototype, self.proto2concept, and self.attention_layer. Thank you very much for sharing. I apologize for bothering you multiple times！

[szzexpoi]

I see. In that case, I may not be able to help. Those weights are not used in the final models; thus, I did not save them in Google Drive. I graduated a year ago and can no longer access the previous lab servers (those weights are probably wiped out as well).

[ProudZhangXiaolong]

Anyway, thanks a million!