Hello everyone,
\nI'm having a hard time to understand how to align train labels with train_player_tracking with the video, since each video has approximately 711 frames. How do I align these data with the video?
\nMy question is how to know each frame will take which row in these csv files.
Thank you!
\nany help is much apppreciated
Hello everyone,
\nI'm having a hard time to understand how to align train labels with train_player_tracking with the video, since each video has approximately 711 frames. How do I align these data with the video?
\nMy question is how to know each frame will take which row in these csv files.
Thank you!
\nany help is much apppreciated
@naifalkhunaizi3 - A few things to keep in mind about the two data sources:
\nBecause of the differences in sample rate it's impossible to perfectly line up the two datasets. One approach is to use the metadata file which gives the approximate start_time of each video, then calculate the approximate timestamp of each frame to merge with the tracking data.
You can check the join_helmets_contact function in the starter notebook here: https://www.kaggle.com/code/robikscube/nfl-player-contact-detection-getting-started
You can also look at solutions from prior years to see how others approached merging this data.
\nHope that helps. Good luck!
", "rawMarkdown": "naifalkhunaizi3 - A few things to keep in mind about the two data sources:\n- Tracking data is collected at 10Hz every 0.1 second should have a datapoint.\n- Video data is collected at 59.94Hz.\n- Submissions are at the tracking data 10Hz.\n\nBecause of the differences in sample rate it's impossible to perfectly line up the two datasets. One approach is to use the `metadata` file which gives the approximate `start_time` of each video, then calculate the approximate timestamp of each frame to merge with the tracking data.\n\nYou can check the `join_helmets_contact` function in the starter notebook here: https://www.kaggle.com/code/robikscube/nfl-player-contact-detection-getting-started\n\nYou can also look at solutions from prior years to see how others approached merging this data.\n\nHope that helps. Good luck!", "votes": null }, { "id": "2063533", "postDate": "12/13/2022 03:28:13", "content": "Can we directly convert between step and frame? I currently use frame = int(step/10*59.95+5*59.95) instead of datetime. Do you think it's right? Thank you.
@zzy990106 I can't say which works best because I don't know. You might want to try different methods and validate if they make a difference. Your calculation looks okay but note you truncate the frame when using int
>>> for step in range(10):\n>>> frame_ = step/10*59.94+5*59.94\n>>> frame = int(frame_)\n>>> print(step, frame, frame_)\n\n0 299 299.7\n1 305 305.69399999999996\n2 311 311.688\n3 317 317.682\n4 323 323.676\n5 329 329.66999999999996\n6 335 335.664\n7 341 341.658\n8 347 347.652\n9 353 353.64599999999996\nTo any that are wondering how this equation works: We know the start of the play (step 0) for Sideline/Endzone videos is roughly 5 seconds into the video and the video is sampled at roughly 59.94Hz. So the frame of step 0 should be 5*59.94 (frame 299). Every other step is an increment of 0.1 seconds and can be computed by adding (step/10*59.94).
Note the video is sampled at 59.94Hz not 59.95Hz. Practically this shouldn't change anything. There was a typo in the data description now fixed.
", "rawMarkdown": "zzy990106 I can't say which works best because I don't know. You might want to try different methods and validate if they make a difference. Your calculation looks okay but note you truncate the frame when using `int`\n\n```\n>>> for step in range(10):\n>>> frame_ = step/10*59.94+5*59.94\n>>> frame = int(frame_)\n>>> print(step, frame, frame_)\n\n0 299 299.7\n1 305 305.69399999999996\n2 311 311.688\n3 317 317.682\n4 323 323.676\n5 329 329.66999999999996\n6 335 335.664\n7 341 341.658\n8 347 347.652\n9 353 353.64599999999996\n```\n\nTo any that are wondering how this equation works: We know the start of the play (step 0) for Sideline/Endzone videos is roughly 5 seconds into the video and the video is sampled at roughly `59.94Hz`. So the frame of step 0 should be 5\\*59.94 (frame 299). Every other step is an increment of 0.1 seconds and can be computed by adding (step/10*59.94).\n\nNote the video is sampled at 59.94Hz not 59.95Hz. Practically this shouldn't change anything. There was a typo in the data description now fixed.", "votes": null }, { "id": "2064191", "postDate": "12/13/2022 15:12:45", "content": "Thank you for detailed explanation. Helps a lot.
", "rawMarkdown": "Thank you for detailed explanation. Helps a lot.", "votes": null } ], "comments": [ { "id": 2063356, "author_name": "robikscube", "author_url": "", "post_date": "12/12/2022 20:14:37", "content": "@naifalkhunaizi3 - A few things to keep in mind about the two data sources:
\nBecause of the differences in sample rate it's impossible to perfectly line up the two datasets. One approach is to use the metadata file which gives the approximate start_time of each video, then calculate the approximate timestamp of each frame to merge with the tracking data.
You can check the join_helmets_contact function in the starter notebook here: https://www.kaggle.com/code/robikscube/nfl-player-contact-detection-getting-started
You can also look at solutions from prior years to see how others approached merging this data.
\nHope that helps. Good luck!
", "votes": null, "replies": [ { "id": 2063533, "author_name": "zzy990106", "author_url": "", "post_date": "12/13/2022 03:28:13", "content": "Can we directly convert between step and frame? I currently use frame = int(step/10*59.95+5*59.95) instead of datetime. Do you think it's right? Thank you.
@zzy990106 I can't say which works best because I don't know. You might want to try different methods and validate if they make a difference. Your calculation looks okay but note you truncate the frame when using int
>>> for step in range(10):\n>>> frame_ = step/10*59.94+5*59.94\n>>> frame = int(frame_)\n>>> print(step, frame, frame_)\n\n0 299 299.7\n1 305 305.69399999999996\n2 311 311.688\n3 317 317.682\n4 323 323.676\n5 329 329.66999999999996\n6 335 335.664\n7 341 341.658\n8 347 347.652\n9 353 353.64599999999996\nTo any that are wondering how this equation works: We know the start of the play (step 0) for Sideline/Endzone videos is roughly 5 seconds into the video and the video is sampled at roughly 59.94Hz. So the frame of step 0 should be 5*59.94 (frame 299). Every other step is an increment of 0.1 seconds and can be computed by adding (step/10*59.94).
Note the video is sampled at 59.94Hz not 59.95Hz. Practically this shouldn't change anything. There was a typo in the data description now fixed.
", "votes": null, "replies": [] }, { "id": 2064191, "author_name": "zzy990106", "author_url": "", "post_date": "12/13/2022 15:12:45", "content": "Thank you for detailed explanation. Helps a lot.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2061777": "Hello everyone, \n\nI'm having a hard time to understand how to align train labels with train_player_tracking with the video, since each video has approximately 711 frames. How do I align these data with the video? \nMy question is how to know each frame will take which row in these csv files. \n\nThank you! \nany help is much apppreciated", "2063356": "naifalkhunaizi3 - A few things to keep in mind about the two data sources:\n- Tracking data is collected at 10Hz every 0.1 second should have a datapoint.\n- Video data is collected at 59.94Hz.\n- Submissions are at the tracking data 10Hz.\n\nBecause of the differences in sample rate it's impossible to perfectly line up the two datasets. One approach is to use the `metadata` file which gives the approximate `start_time` of each video, then calculate the approximate timestamp of each frame to merge with the tracking data.\n\nYou can check the `join_helmets_contact` function in the starter notebook here: https://www.kaggle.com/code/robikscube/nfl-player-contact-detection-getting-started\n\nYou can also look at solutions from prior years to see how others approached merging this data.\n\nHope that helps. Good luck!", "2063533": "Can we directly convert between step and frame? I currently use `frame = int(step/10*59.95+5*59.95)` instead of datetime. Do you think it's right? Thank you.", "2064184": "zzy990106 I can't say which works best because I don't know. You might want to try different methods and validate if they make a difference. Your calculation looks okay but note you truncate the frame when using `int`\n\n```\n>>> for step in range(10):\n>>> frame_ = step/10*59.94+5*59.94\n>>> frame = int(frame_)\n>>> print(step, frame, frame_)\n\n0 299 299.7\n1 305 305.69399999999996\n2 311 311.688\n3 317 317.682\n4 323 323.676\n5 329 329.66999999999996\n6 335 335.664\n7 341 341.658\n8 347 347.652\n9 353 353.64599999999996\n```\n\nTo any that are wondering how this equation works: We know the start of the play (step 0) for Sideline/Endzone videos is roughly 5 seconds into the video and the video is sampled at roughly `59.94Hz`. So the frame of step 0 should be 5\\*59.94 (frame 299). Every other step is an increment of 0.1 seconds and can be computed by adding (step/10*59.94).\n\nNote the video is sampled at 59.94Hz not 59.95Hz. Practically this shouldn't change anything. There was a typo in the data description now fixed.", "2064191": "Thank you for detailed explanation. Helps a lot." }, "source": "meta" }