china-uk-zsl / zs-f-vqa Goto Github PK

你好，我在阅读代码过程中发现你所检索的Knowledge Base好像不是包含 193,449个fact的Knowledge Base，而是每个问题所对应的fact组成的Knowledge Base, 我想确认一下这样是可以的吗

Dear scholar, Could you provide your fvqa_36.hdf5 and fvqa36_imgid2idx.pkl?

where can I set the root path?

Hello, I try to reproduce your code, but I don’t know how to generate fvqa-resnet-14x14.h5. Can you share this file with me, thank you very much

Thanks for your paper and shared codes!
I'm new at this area . I wonder how to evaluate my own data(input an image and a question , then get an answer) by establishing an API

Thanks for sharing the code, and I would like to know how to train this model on two GPUs?

Great work.

By the way, when i run joint_test.py with another gpu you used, a run time error occurs.

python joint_test.py --gpu_id 0 --exp_name fusion_prediction --ZSL 0 --exp_id rel3_fact10data_3score_10 --data_choice 3 --top_rel 3 --top_fact 10 --soft_score 10 --mrr 1

Maybe you can simply change line 336 in joint_test.py with below link
https://stackoverflow.com/questions/53186736/

Hi, now I can complet the eval, but I also want to get the predicted answer file. Looking forward to your reply！

请问图片直接下载您给的FVQA放在 data/KG_VQA/fvqa/exp_data/images/images文件下，可以直接运行代码吗？

个人感觉您写的README逻辑不够清晰，您能写的清楚点嘛？

Dear Prof:
Could you please tell me how to get the feature file for the BUTD or the BAN model?

你好，我现在在使用自己的数据集进行训练，我按照您项目中的json文件标注的问题答案以及对应的fact。
但是我想知道数据检查部分是什么用途，项目中有很多filter文件，我自己的数据集需要怎么做呢