Question for the organizers: For the train data - and the test data - are the frames per second consistent across all data? Consistent for a single participant (based on their hardware?) but not consistent across all data?
\nIf consistent, what is it?
\nThe data tab description says that 0.5 seconds before and after record button are captured. On the other hand, the average number of frames in total is pretty low (<40), and the minimum per sign ranges from just 2-6 frames, implying that frames per second isn't consistent. (Or just that in rare cases frames might not be captured successfully.)
\nGiven the direct relevance to the problem, and how little visibility we have, it would really help to be given a little more understanding into how time vs frames will manifest across many observations from a small or large number of users?
", "messages": [ { "id": "2160568", "postDate": "02/26/2023 19:14:18", "content": "Question for the organizers: For the train data - and the test data - are the frames per second consistent across all data? Consistent for a single participant (based on their hardware?) but not consistent across all data?
\nIf consistent, what is it?
\nThe data tab description says that 0.5 seconds before and after record button are captured. On the other hand, the average number of frames in total is pretty low (<40), and the minimum per sign ranges from just 2-6 frames, implying that frames per second isn't consistent. (Or just that in rare cases frames might not be captured successfully.)
\nGiven the direct relevance to the problem, and how little visibility we have, it would really help to be given a little more understanding into how time vs frames will manifest across many observations from a small or large number of users?
", "rawMarkdown": "Question for the organizers: For the train data - and the test data - are the frames per second consistent across all data? Consistent for a single participant (based on their hardware?) but not consistent across all data?\n\nIf consistent, what is it?\n\nThe data tab description says that 0.5 seconds before and after record button are captured. On the other hand, the average number of frames in total is pretty low (<40), and the minimum per sign ranges from just 2-6 frames, implying that frames per second isn't consistent. (Or just that in rare cases frames might not be captured successfully.)\n\nGiven the direct relevance to the problem, and how little visibility we have, it would really help to be given a little more understanding into how time vs frames will manifest across many observations from a small or large number of users?", "votes": null }, { "id": "2160641", "postDate": "02/26/2023 20:17:36", "content": "Mediapipe can lose tracking, so there not a consistent number of frames per second. It is something the Popsign game will have to accept.
", "rawMarkdown": "Mediapipe can lose tracking, so there not a consistent number of frames per second. It is something the Popsign game will have to accept.", "votes": null }, { "id": "2160654", "postDate": "02/26/2023 20:35:14", "content": "If it doesn't \"lose tracking\", then would it be consistent? Or still vary based on hardware or software or other factors?
", "rawMarkdown": "If it doesn't \"lose tracking\", then would it be consistent? Or still vary based on hardware or software or other factors?", "votes": null }, { "id": "2162469", "postDate": "02/28/2023 08:11:56", "content": "Data seems to had been collected using some-kind of crowd-sourcing by having community record themselves while holding a button with one hand and performing the sign with the other hand.
\nThe participants used their own phone's selfie camera for doing the recording.
\nThis means that depending on light condition the FPS of the camera would vary to adjust - low light would mean less FPS, I wouldn't discard the possibility of 1 FPS.
\nHaving more or enough light would result in recordings at the target video output FPS - which could be somewhere between 24-60 FPS.
\nSo 30 seconds video could be 30 frames or 1800 frames …
", "rawMarkdown": "Data seems to had been collected using some-kind of crowd-sourcing by having community record themselves while holding a button with one hand and performing the sign with the other hand. \n\nThe participants used their own phone's selfie camera for doing the recording. \n\nThis means that depending on light condition the FPS of the camera would vary to adjust - low light would mean less FPS, I wouldn't discard the possibility of 1 FPS. \n\nHaving more or enough light would result in recordings at the target video output FPS - which could be somewhere between 24-60 FPS.\n\nSo 30 seconds video could be 30 frames or 1800 frames ...", "votes": null }, { "id": "2163403", "postDate": "02/28/2023 19:25:51", "content": "Okay, thanks, that makes sense. Which means it can be all over the place, especially if mediapipe also drops a lot of frames.
\nStill, given the request for 21 people to each take a bunch of videos, one would normally expect good lighting and thus to hit the target video output >70% of the time. And one would expect that to usually be at least 0.5s before, at least 0.25-0.5sec during, 0.5s after, which should have a lower bound of around 30-36 frames. But with a mean of ~38, I haven't checked the 25%-ile or the median, but it seems like there are far more videos with <20 or <30 frames than would be expected if mediapipe only rarely drops frames, and if most videos have enough light and run at normal FPS of 24+.
", "rawMarkdown": "Okay, thanks, that makes sense. Which means it can be all over the place, especially if mediapipe also drops a lot of frames.\n\nStill, given the request for 21 people to each take a bunch of videos, one would normally expect good lighting and thus to hit the target video output >70% of the time. And one would expect that to usually be at least 0.5s before, at least 0.25-0.5sec during, 0.5s after, which should have a lower bound of around 30-36 frames. But with a mean of ~38, I haven't checked the 25%-ile or the median, but it seems like there are far more videos with <20 or <30 frames than would be expected if mediapipe only rarely drops frames, and if most videos have enough light and run at normal FPS of 24+.", "votes": null }, { "id": "2191678", "postDate": "03/22/2023 06:08:52", "content": "\n\n\nData seems to had been collected using some-kind of crowd-sourcing by having community record themselves while holding a button with one hand and performing the sign with the other hand.
\n
\n The participants used their own phone's selfie camera for doing the recording.
\nI found this description in data card tab.
\n", "rawMarkdown": "profetul \n\n> Data seems to had been collected using some-kind of crowd-sourcing by having community record themselves while holding a button with one hand and performing the sign with the other hand.\n> The participants used their own phone's selfie camera for doing the recording.\n\n~~Do you have any information source? Or, is this just your guess?~~\nI found this description in data card tab.\n\n> Signers who communicate using American Sign Language as their primary language were recruited from across the United States. They were shipped a Pixel 4a smartphone with an installed collection app. The app prompted the signer with the concept in English to sign, randomly selected from the 250-sign vocabulary. Signers pressed and held an on-screen button on the phone to record video while signing each concept, releasing the button after each sign. The video of the sign is extracted with a buffer 0.5 seconds before the press of the button and 0.5 seconds after the release of the button.", "votes": null }, { "id": "2191714", "postDate": "03/22/2023 06:40:20", "content": "\nSigners who communicate using American Sign Language as their primary language were recruited from across the United States. They were shipped a Pixel 4a smartphone with an installed collection app. The app prompted the signer with the concept in English to sign, randomly selected from the 250-sign vocabulary. Signers pressed and held an on-screen button on the phone to record video while signing each concept, releasing the button after each sign. The video of the sign is extracted with a buffer 0.5 seconds before the press of the button and 0.5 seconds after the release of the button.
\n
\n\nThis means that depending on light condition the FPS of the camera would vary to adjust - low light would mean less FPS, I wouldn't discard the possibility of 1 FPS.
\n
Could you show us the information source of Google Pixel 4a's camera's FPS differ based on the light conditions?
", "rawMarkdown": "profetul \n\n> This means that depending on light condition the FPS of the camera would vary to adjust - low light would mean less FPS, I wouldn't discard the possibility of 1 FPS.\n\nCould you show us the information source of Google Pixel 4a's camera's FPS differ based on the light conditions?", "votes": null } ], "comments": [ { "id": 2160641, "author_name": "thadstarner", "author_url": "", "post_date": "02/26/2023 20:17:36", "content": "Mediapipe can lose tracking, so there not a consistent number of frames per second. It is something the Popsign game will have to accept.
", "votes": null, "replies": [ { "id": 2160654, "author_name": "roberthatch", "author_url": "", "post_date": "02/26/2023 20:35:14", "content": "If it doesn't \"lose tracking\", then would it be consistent? Or still vary based on hardware or software or other factors?
", "votes": null, "replies": [] } ] }, { "id": 2162469, "author_name": "profetul", "author_url": "", "post_date": "02/28/2023 08:11:56", "content": "Data seems to had been collected using some-kind of crowd-sourcing by having community record themselves while holding a button with one hand and performing the sign with the other hand.
\nThe participants used their own phone's selfie camera for doing the recording.
\nThis means that depending on light condition the FPS of the camera would vary to adjust - low light would mean less FPS, I wouldn't discard the possibility of 1 FPS.
\nHaving more or enough light would result in recordings at the target video output FPS - which could be somewhere between 24-60 FPS.
\nSo 30 seconds video could be 30 frames or 1800 frames …
", "votes": null, "replies": [ { "id": 2163403, "author_name": "roberthatch", "author_url": "", "post_date": "02/28/2023 19:25:51", "content": "Okay, thanks, that makes sense. Which means it can be all over the place, especially if mediapipe also drops a lot of frames.
\nStill, given the request for 21 people to each take a bunch of videos, one would normally expect good lighting and thus to hit the target video output >70% of the time. And one would expect that to usually be at least 0.5s before, at least 0.25-0.5sec during, 0.5s after, which should have a lower bound of around 30-36 frames. But with a mean of ~38, I haven't checked the 25%-ile or the median, but it seems like there are far more videos with <20 or <30 frames than would be expected if mediapipe only rarely drops frames, and if most videos have enough light and run at normal FPS of 24+.
", "votes": null, "replies": [] }, { "id": 2191678, "author_name": "tatamikenn", "author_url": "", "post_date": "03/22/2023 06:08:52", "content": "\n\n\nData seems to had been collected using some-kind of crowd-sourcing by having community record themselves while holding a button with one hand and performing the sign with the other hand.
\n
\n The participants used their own phone's selfie camera for doing the recording.
\nI found this description in data card tab.
\n", "votes": null, "replies": [ { "id": 2191714, "author_name": "tatamikenn", "author_url": "", "post_date": "03/22/2023 06:40:20", "content": "\nSigners who communicate using American Sign Language as their primary language were recruited from across the United States. They were shipped a Pixel 4a smartphone with an installed collection app. The app prompted the signer with the concept in English to sign, randomly selected from the 250-sign vocabulary. Signers pressed and held an on-screen button on the phone to record video while signing each concept, releasing the button after each sign. The video of the sign is extracted with a buffer 0.5 seconds before the press of the button and 0.5 seconds after the release of the button.
\n
\n\nThis means that depending on light condition the FPS of the camera would vary to adjust - low light would mean less FPS, I wouldn't discard the possibility of 1 FPS.
\n
Could you show us the information source of Google Pixel 4a's camera's FPS differ based on the light conditions?
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2160568": "Question for the organizers: For the train data - and the test data - are the frames per second consistent across all data? Consistent for a single participant (based on their hardware?) but not consistent across all data?\n\nIf consistent, what is it?\n\nThe data tab description says that 0.5 seconds before and after record button are captured. On the other hand, the average number of frames in total is pretty low (<40), and the minimum per sign ranges from just 2-6 frames, implying that frames per second isn't consistent. (Or just that in rare cases frames might not be captured successfully.)\n\nGiven the direct relevance to the problem, and how little visibility we have, it would really help to be given a little more understanding into how time vs frames will manifest across many observations from a small or large number of users?", "2160641": "Mediapipe can lose tracking, so there not a consistent number of frames per second. It is something the Popsign game will have to accept.", "2160654": "If it doesn't \"lose tracking\", then would it be consistent? Or still vary based on hardware or software or other factors?", "2162469": "Data seems to had been collected using some-kind of crowd-sourcing by having community record themselves while holding a button with one hand and performing the sign with the other hand. \n\nThe participants used their own phone's selfie camera for doing the recording. \n\nThis means that depending on light condition the FPS of the camera would vary to adjust - low light would mean less FPS, I wouldn't discard the possibility of 1 FPS. \n\nHaving more or enough light would result in recordings at the target video output FPS - which could be somewhere between 24-60 FPS.\n\nSo 30 seconds video could be 30 frames or 1800 frames ...", "2163403": "Okay, thanks, that makes sense. Which means it can be all over the place, especially if mediapipe also drops a lot of frames.\n\nStill, given the request for 21 people to each take a bunch of videos, one would normally expect good lighting and thus to hit the target video output >70% of the time. And one would expect that to usually be at least 0.5s before, at least 0.25-0.5sec during, 0.5s after, which should have a lower bound of around 30-36 frames. But with a mean of ~38, I haven't checked the 25%-ile or the median, but it seems like there are far more videos with <20 or <30 frames than would be expected if mediapipe only rarely drops frames, and if most videos have enough light and run at normal FPS of 24+.", "2191678": "profetul \n\n> Data seems to had been collected using some-kind of crowd-sourcing by having community record themselves while holding a button with one hand and performing the sign with the other hand.\n> The participants used their own phone's selfie camera for doing the recording.\n\n~~Do you have any information source? Or, is this just your guess?~~\nI found this description in data card tab.\n\n> Signers who communicate using American Sign Language as their primary language were recruited from across the United States. They were shipped a Pixel 4a smartphone with an installed collection app. The app prompted the signer with the concept in English to sign, randomly selected from the 250-sign vocabulary. Signers pressed and held an on-screen button on the phone to record video while signing each concept, releasing the button after each sign. The video of the sign is extracted with a buffer 0.5 seconds before the press of the button and 0.5 seconds after the release of the button.", "2191714": "profetul \n\n> This means that depending on light condition the FPS of the camera would vary to adjust - low light would mean less FPS, I wouldn't discard the possibility of 1 FPS.\n\nCould you show us the information source of Google Pixel 4a's camera's FPS differ based on the light conditions?" }, "source": "meta" }