wujie1010 / temporally-language-grounding Goto Github PK

thanks for your sharing!
I run your code and get the results,but the performance is much worse than yours.
I have no idea about this,could you give me some advice?

Thanks for your sharing!

I encountered a problem when training MAC: Dataloader is extremely slow to load data even if I increase the num_work value. In the case of num_work equal to 16, a batch of data takes about 8 minutes to load. So I want to ask if you have this problem or if there is something wrong with my settings.

I noticed that the number of uncompressed visual feature and visual activity concepts tar files is huge (the former is about 940,000 and the latter is about 640,000).

I try to run your code, but I failed because of the pickle.load.

self.clip_sentence_pairs_iou_all = pickle.load(open("./Dataset/Charades/ref_info/charades_rl_train_feature.pkl"))

the the error is

TypeError: a bytes-like object is required, not 'str'

then I try

pickle.load(open("./Dataset/Charades/ref_info/charades_rl_train_feature.pkl",'rb'))

but the error is

UnicodeDecodeError: 'ascii' codec can't decode byte 0x81 in position 0: ordinal not in range(128)

I also tried several methods, but none of them working. Such as add encoding/decoding, import sys ...... I also try to use python2.7 load the data, but I get the same errors.

Can you release the code about preprocessing the visual / sentence feature?
I want to know about how you get the visual feature (ResNet or others?) and what is the exact meaning of the sentence feature files.

hi WuJie,Sorry to bother you. I was reading your code to reproduce the A2C model and I was wondering how you dealt with the dataset. How to get ref_info folder and charades_rl_train_feature.pkl. Maybe my knowledge level is not enough, I hope to get your advice, thank you very much!

@WuJie1010 Thanks for the nice sharing!
Could you please tell me where is the data under the directory train_softmax_dir?

1. In the code, you didn't add Location/IoU loss, what you do is just create the variables without assigning any value. Actually, if I try to revise the loss function into correct code. I find the performance will decrease largely.

2. The policy loss function is always negative and what we want to do is minimum this loss rather than make it always increase. I was wondering the reason is reward value is always lower than state value(According to the A2C formula) or this is the bug of the code.