questions about the dataset

Question

questions about the dataset

Closed this issue 8 months ago · 9 comments

Hello, I'm not very clear about some parts of this dataset. Can you help me?
In this project, we need to run extract_ frames. py to extract frames from video of bdd100k dataset. I can't judge whether I am right or wrong in this step.
Firstly, The problem is that the frame I extracted is rotated 180 ° clockwise. Then, the number of frames extracted here is not exactly the same as that given in file train_day.list. For example, about video 07f1933b-8c2284ed.mov, I extracted 1205 pictures, but it is indicated in the train_day.list that the video has 1202 frames in total.
Did my operation cause an error? Can you give an example of this process? We look forward to your reply. Thank you very much.

Answer 1 · 2022-12-02T15:17:53.000Z

And I want to ask, from Table 3 in the paper, [top-1, top-5, top-10, top-1%] = [15.2, 27.2, 33.8, 91.9] is the result of the Step-1 of the hierarchical approach?

Answer 2 · 2022-12-02T15:54:54.000Z

Hello zexianghui Yes, the frames were extracted using extract_ frames. py however I am not sure why the frames would rotate. There were some differences in the pixel sizes/orientation of some videos which caused the issue. If you are seeing the rotation with only some of the videos that may be the issue. It is possible that an older file got shared. If it is fixed at your end please let me know what was the problem. Some of the frames (in 2-3 videos) got corrupt during my extraction and I probably deleted/removed them. I would suggest to update/create the number of frames using train_day.list as a list of video folders to iterate over.

…

On Fri, Dec 2, 2022 at 6:22 AM zexianghui ***@***.***> wrote: Hello, I'm not very clear about some parts of this dataset. Can you help me? In this project, we need to run extract_ frames. py to extract frames from video of bdd100k dataset. I can't judge whether I am right or wrong in this step. Firstly, The problem is that the frame I extracted is rotated 180 ° clockwise. Then, the number of frames extracted here is not exactly the same as that given in file train_day.list. For example, about video 07f1933b-8c2284ed.mov, I extracted 1205 pictures, but it is indicated in the train_day.list that the video has 1202 frames in total. Did my operation cause an error? Can you give an example of this process? We look forward to your reply. Thank you very much. — Reply to this email directly, view it on GitHub <#4>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQIWMAYTIXDRDH4ZY6Q5Q43WLHLYPANCNFSM6AAAAAASRZ455E> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

Answer 3 · 2022-12-02T15:56:27.000Z

Yes, this is correct.

…

On Fri, Dec 2, 2022 at 10:18 AM zexianghui ***@***.***> wrote: And I want to ask, from Table 3 in the paper, [top-1, top-5, top-10, top-1%] = [15.2, 27.2, 33.8, 91.9] is the result of the Step-1 of the hierarchical approach? — Reply to this email directly, view it on GitHub <#4 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQIWMA33QPN5AXJJQVSYKOLWLIHKZANCNFSM6AAAAAASRZ455E> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

Answer 4 · 2022-12-03T03:36:00.000Z

Thanks for your reply.
I run extract_ frames.py. All images I got are inverted. For example, about 0024b742-83709bd4.mov in the training set. The image I got is as follows. Is this the frame picture used in your experiment?

Thanks again for your reply.

Answer 5 · 2022-12-03T16:38:37.000Z

My frames were all straight without rotation. Maybe there is some error at your end. See, if some transformation got uncommented.

…

On Fri, Dec 2, 2022, 10:36 PM zexianghui ***@***.***> wrote: Thanks for your reply. I run extract_ frames.py. All images I got are inverted. For example, about 0024b742-83709bd4.mov in the training set. The image I got is as follows. Is this the frame picture used in your experiment? [image: image] <https://user-images.githubusercontent.com/57891691/205420263-ea6864b0-d266-40af-9c1e-d62db3906887.png> Thanks again for your reply. — Reply to this email directly, view it on GitHub <#4 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQIWMAYOEOYE6REI334SWX3WLK52VANCNFSM6AAAAAASRZ455E> . You are receiving this because you commented.Message ID: ***@***.***>

Answer 6 · 2022-12-05T03:21:34.000Z

This is the result of running extract_ frames. py directly. Would you please check the code?
Can you upload a result of extract_ frames. py for a video?Thanks very much.

Answer 7 · 2022-12-12T14:07:07.000Z

This is the result of running extract_ frames. py directly. Would you please check the code? Can you upload a result of extract_ frames. py for a video?Thanks very much.

Same problem! I've spent a lot of time understanding the code. Were you able to fix it? @svyas23 Thank you for your time.

Answer 8 · 2024-01-26T16:47:05.000Z

And I want to ask, from Table 3 in the paper, [top-1, top-5, top-10, top-1%] = [15.2, 27.2, 33.8, 91.9] is the result of the Step-1 of the hierarchical approach?

Yes that's true

Answer 9 · 2024-08-14T02:46:47.000Z

I have a similar concern regarding the dataset processing code "extract_ frames. py". The original code I pulled from your repository might have some bugs and cannot be run directly. After the modification by myself, the same situation as others occurred: the extracted image was incorrect. Could you please take a look and help resolve them? Thank you!