![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1996737%2F805ec5b2fc3ac1528ddeae246516339d%2Fframe_0.png?generation=1679503327946121&alt=media\")
Hi, guys.
\nThank's to community public notebooks, I managed to create my own sample visualizer. I've split the sample into 4 parts: hands, face, and pose. And I saw a lot of poses, with feet presented. I've tried to create a picture with my smartphone camera and having a face in the frame I wasn't able to capture anything below the hips.
\nBut in the dataset full pose is presented.
\nAccording to the data description: \"Signers pressed and held an on-screen button on the phone to record video while signing each concept, releasing the button after each sign.\". With that said, I have no other idea except enormously long hands.
Did I miss something important? Any thoughts on this? Thank you in advance!
\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1996737%2F3b96def0b58dbd803632e7d96b7954c0%2Fframe_0.png?generation=1679503362575484&alt=media\")
Hi, guys.
\nThank's to community public notebooks, I managed to create my own sample visualizer. I've split the sample into 4 parts: hands, face, and pose. And I saw a lot of poses, with feet presented. I've tried to create a picture with my smartphone camera and having a face in the frame I wasn't able to capture anything below the hips.
\nBut in the dataset full pose is presented.
\nAccording to the data description: \"Signers pressed and held an on-screen button on the phone to record video while signing each concept, releasing the button after each sign.\". With that said, I have no other idea except enormously long hands.
Did I miss something important? Any thoughts on this? Thank you in advance!
\n
Mediapipe always gives values for all pose and face keypoints. That does not mean all of them were visible in the frame, the ones of out of frame are just extrapolated/guessed.
", "rawMarkdown": "Mediapipe always gives values for all pose and face keypoints. That does not mean all of them were visible in the frame, the ones of out of frame are just extrapolated/guessed.", "votes": null }, { "id": "2192576", "postDate": "03/22/2023 18:27:16", "content": "Thank you so much. That really makes sense!
\nYou mentioned, that mediapipe always gives values for all pose and face landmarks. I may be wrong, but I saw a sample, where one of the frames of the face was missed. Do you mean, that once the pose is presented, all the landmarks are given (maybe with extrapolation), but there is a possibility of missed pose or face?
I haven't checked all the samples but as far as I could tell only hands seem to be missing from time to time (well actually they are missing fairly often).
", "rawMarkdown": "I haven't checked all the samples but as far as I could tell only hands seem to be missing from time to time (well actually they are missing fairly often).", "votes": null }, { "id": "2193344", "postDate": "03/23/2023 08:21:00", "content": "That makes sense since the on-screen button is supposed to be pressed during recording. Sometimes, a hand is presented but a few frames are missed. I feel the same may happen with the face as well (e.g. jerky movement of face or smartphone)
", "rawMarkdown": "That makes sense since the on-screen button is supposed to be pressed during recording. Sometimes, a hand is presented but a few frames are missed. I feel the same may happen with the face as well (e.g. jerky movement of face or smartphone)", "votes": null }, { "id": "2193405", "postDate": "03/23/2023 09:22:10", "content": "Yeah, there are some weird leg movements in the dataset too. Perhaps real data looks something like this: https://www.kaggle.com/competitions/asl-signs/discussion/391265#2183164
", "rawMarkdown": "Yeah, there are some weird leg movements in the dataset too. Perhaps real data looks something like this: https://www.kaggle.com/competitions/asl-signs/discussion/391265#2183164", "votes": null }, { "id": "2193450", "postDate": "03/23/2023 10:00:41", "content": "I am pretty sure we should ignore legs
", "rawMarkdown": "I am pretty sure we should ignore legs", "votes": null } ], "comments": [ { "id": 2192493, "author_name": "mxbonn", "author_url": "", "post_date": "03/22/2023 17:26:53", "content": "Mediapipe always gives values for all pose and face keypoints. That does not mean all of them were visible in the frame, the ones of out of frame are just extrapolated/guessed.
", "votes": null, "replies": [ { "id": 2192576, "author_name": "meowmeowmeowmeowmeow", "author_url": "", "post_date": "03/22/2023 18:27:16", "content": "Thank you so much. That really makes sense!
\nYou mentioned, that mediapipe always gives values for all pose and face landmarks. I may be wrong, but I saw a sample, where one of the frames of the face was missed. Do you mean, that once the pose is presented, all the landmarks are given (maybe with extrapolation), but there is a possibility of missed pose or face?
I haven't checked all the samples but as far as I could tell only hands seem to be missing from time to time (well actually they are missing fairly often).
", "votes": null, "replies": [ { "id": 2193344, "author_name": "meowmeowmeowmeowmeow", "author_url": "", "post_date": "03/23/2023 08:21:00", "content": "That makes sense since the on-screen button is supposed to be pressed during recording. Sometimes, a hand is presented but a few frames are missed. I feel the same may happen with the face as well (e.g. jerky movement of face or smartphone)
", "votes": null, "replies": [] } ] } ] }, { "id": 2193405, "author_name": "vbogach", "author_url": "", "post_date": "03/23/2023 09:22:10", "content": "Yeah, there are some weird leg movements in the dataset too. Perhaps real data looks something like this: https://www.kaggle.com/competitions/asl-signs/discussion/391265#2183164
", "votes": null, "replies": [ { "id": 2193450, "author_name": "meowmeowmeowmeowmeow", "author_url": "", "post_date": "03/23/2023 10:00:41", "content": "I am pretty sure we should ignore legs
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2192429": "Hi, guys.\nThank's to community public notebooks, I managed to create my own sample visualizer. I've split the sample into 4 parts: hands, face, and pose. And I saw a lot of poses, with feet presented. I've tried to create a picture with my smartphone camera and having a face in the frame I wasn't able to capture anything below the hips.\nBut in the dataset full pose is presented. \nAccording to the data description: \"Signers pressed and held an on-screen button on the phone to record video while signing each concept, releasing the button after each sign.\". With that said, I have no other idea except enormously long hands. \n\nDid I miss something important? Any thoughts on this? Thank you in advance!\n\n\n\n", "2192493": "Mediapipe always gives values for all pose and face keypoints. That does not mean all of them were visible in the frame, the ones of out of frame are just extrapolated/guessed.", "2192576": "Thank you so much. That really makes sense! \nYou mentioned, that mediapipe always gives values for all pose and face landmarks. I may be wrong, but I saw a sample, where one of the frames of the face was missed. Do you mean, that once the pose is presented, all the landmarks are given (maybe with extrapolation), but there is a possibility of missed pose or face?", "2192647": "I haven't checked all the samples but as far as I could tell only hands seem to be missing from time to time (well actually they are missing fairly often).", "2193344": "That makes sense since the on-screen button is supposed to be pressed during recording. Sometimes, a hand is presented but a few frames are missed. I feel the same may happen with the face as well (e.g. jerky movement of face or smartphone)", "2193405": "Yeah, there are some weird leg movements in the dataset too. Perhaps real data looks something like this: https://www.kaggle.com/competitions/asl-signs/discussion/391265#2183164", "2193450": "I am pretty sure we should ignore legs" }, "source": "meta" }