incorrect data structures for your TSA

[tzzcl]

Hi, Thank you for your amazing work on action quality assessment!

I try to run your TSA with the downloaded dataset, but it will post the following error:

FineDiving_Pair.py", line 69, in load_video
    start_frame = int(image_list[0].split("/")[-1][:-4])
IndexError: list index out of range

When I check the organization of the data, I found that your code seems need two subfolders: video/subpart/*.jpg, while in the downloaded untrimmed frames/extracted untrimmed frames, it only has video/*.jpg. So can you explain how I can run the sample code?

Thank you!

[tzzcl]

Moreover, I've tested your code, the test.sh will output the following results:

[TEST] tIoU_5: 0.893191, tIoU_75: 0.471295 [TEST] correlation: 0.929848, L2: 32.764969, RL2: 0.003006

which is much higher than your results is the paper?
Do I made any mistake or your have updated results?

[xujinglin]

Hi, thanks for your interest in our work.

The procedure segmentation output is of a random nature, which may result in the performance gap in rho and tIoU. Besides, prob_tas_threshold and max_epoch are the pre-defined parameters, and prob_tas_threshold * max_epoch indicates the number of epochs used to guide the procedure segmentation obtaining a good initialization, which also results in a performance gap in tIoU.

[hu-wei-carrzy]

Hi, Thank you for your amazing work on action quality assessment!

I try to run your TSA with the downloaded dataset, but it will post the following error:

FineDiving_Pair.py", line 69, in load_video
    start_frame = int(image_list[0].split("/")[-1][:-4])
IndexError: list index out of range

When I check the organization of the data, I found that your code seems need two subfolders: video/subpart/*.jpg, while in the downloaded untrimmed frames/extracted untrimmed frames, it only has video/*.jpg. So can you explain how I can run the sample code?

Thank you!

I
I have also encountered this problem. How did you solve it