Low R1 performance in the 2nd stage

[chenhao2345]

Thanks for sharing your code. Is it normal to get R1=30 with train_titles.py? After running the score fusion, the title matrix does not improve the video matrix.

[BishmoyPaul]

I am having the same issue. Incidentally, did you have any such error with the fusion score? I was running it on MSVD

Text-to-Video:
>>>  R@1: 30.4 - R@5: 59.7 - R@10: 70.7 - Median R: 3.0 - Mean R: 19.8
Video-to-Text:
>>>  V2T$R@1: 33.1 - V2T$R@5: 60.1 - V2T$R@10: 72.6 - V2T$Median R: 3.0 - V2T$Mean R: 18.4
video_matrix sim matrix size: (27763, 670), (27763, 670)
titles_shot_matrix sim matrix size: (27763, 670), (27763, 670)
Traceback (most recent call last):
  File "/local/Cap4Video/train_titles.py", line 723, in <module>
    fusion_scores()
  File "/local/Cap4Video/sim_matrix/fusion_scores.py", line 13, in fusion_scores
    tv_video_metrics = compute_metrics(video_matrix)
  File "/local/Cap4Video/metrics.py", line 13, in compute_metrics
    ind = sx - d
ValueError: operands could not be broadcast together with shapes (27763,670) (670,1) 

[chenhao2345]

@BishmoyPaul I'm running it on MSRVTT. I have not seen any problems with the fusion score on MSRVTT.

[JosephPai]

Same problem here. @whwu95
I got Rank-1 47.7 in the first stage train_video.py
And Rank-1 around 30 in the second stage train_titles.py.

[JosephPai]

BTW, do you know the purpose of fusion_scores? @chenhao2345

[chenhao2345]

@JosephPai I got similar performance. ~47.5 in stage 1 and 30 in stage 2.

It seems to me that the authors get two similarity scores from stage 1 and stage 2, respectively. Then, they use fusion_scores to fuze the two similarity scores.

[ASENNIU]

I got R@1 45.3 in stage and 29.6 in stage 2, it seems like that the code is to do global matching ?

[shams2023]

I got R@1 45.3 in stage and 29.6 in stage 2, it seems like that the code is to do global matching ?

i think its true

[ASENNIU]

Thanks for sharing your code. And how can I get the score 49 for R@1?

[zef1611]

@chenhao2345 @JosephPai @ASENNIU @BishmoyPaul Hi, can I know your batch size setting and the number of gpus you are using for training stage 1 & stage 2?

[fazliimam]

@zef1611 did you find the batch size, number of gpus and gpu type used in this project?
can anyone please answer this? @chenhao2345 @JosephPai @ASENNIU @BishmoyPaul

[fazliimam]

I am having the same issue. Incidentally, did you have any such error with the fusion score? I was running it on MSVD

Text-to-Video:
>>>  R@1: 30.4 - R@5: 59.7 - R@10: 70.7 - Median R: 3.0 - Mean R: 19.8
Video-to-Text:
>>>  V2T$R@1: 33.1 - V2T$R@5: 60.1 - V2T$R@10: 72.6 - V2T$Median R: 3.0 - V2T$Mean R: 18.4
video_matrix sim matrix size: (27763, 670), (27763, 670)
titles_shot_matrix sim matrix size: (27763, 670), (27763, 670)
Traceback (most recent call last):
  File "/local/Cap4Video/train_titles.py", line 723, in <module>
    fusion_scores()
  File "/local/Cap4Video/sim_matrix/fusion_scores.py", line 13, in fusion_scores
    tv_video_metrics = compute_metrics(video_matrix)
  File "/local/Cap4Video/metrics.py", line 13, in compute_metrics
    ind = sx - d
ValueError: operands could not be broadcast together with shapes (27763,670) (670,1) 

@BishmoyPaul How did you train on the MSVD dataset?
If you were using the co_train_msrvtt.sh script what did you give for --data_path
Could you share the training script