On committing my notebook took around 3.5 hours. But on submitting, notebook timeout occurs. I searched and got to know on submitting our model would be run through 4000 videos (is it including 400 sample test videos or else 4000 + 400 videos?). As 9 hours is the limit, so 8.1 seconds per video for preprocessing + prediction should be the worst case. But my preprocessing itself is exceeding the time limit in which I am just simply capturing 40 frames per video and directly sending it to my model. How can one reduce the capture time?
", "messages": [ { "id": "752136", "postDate": "02/20/2020 18:58:10", "content": "On committing my notebook took around 3.5 hours. But on submitting, notebook timeout occurs. I searched and got to know on submitting our model would be run through 4000 videos (is it including 400 sample test videos or else 4000 + 400 videos?). As 9 hours is the limit, so 8.1 seconds per video for preprocessing + prediction should be the worst case. But my preprocessing itself is exceeding the time limit in which I am just simply capturing 40 frames per video and directly sending it to my model. How can one reduce the capture time?
", "rawMarkdown": "On committing my notebook took around 3.5 hours. But on submitting, notebook timeout occurs. I searched and got to know on submitting our model would be run through 4000 videos (is it including 400 sample test videos or else 4000 + 400 videos?). As 9 hours is the limit, so 8.1 seconds per video for preprocessing + prediction should be the worst case. But my preprocessing itself is exceeding the time limit in which I am just simply capturing 40 frames per video and directly sending it to my model. How can one reduce the capture time?", "votes": null }, { "id": "752195", "postDate": "02/20/2020 19:34:52", "content": "Some things you can try:
\n\nHi, I see you using multi-threading to speed up the kernel, is this effective？\nHow do I know how many threads should I set？\nThank you
", "rawMarkdown": "Hi, I see you using multi-threading to speed up the kernel, is this effective？\nHow do I know how many threads should I set？\nThank you", "votes": null }, { "id": "752441", "postDate": "02/21/2020 02:51:37", "content": "You can query the number of CPUs available and then use that value for splitting work. But it's not really useful for this competition.
", "rawMarkdown": "You can query the number of CPUs available and then use that value for splitting work. But it's not really useful for this competition.", "votes": null }, { "id": "752768", "postDate": "02/21/2020 11:41:54", "content": "I think you get only 2 CPU cores when GPU is enabled, so the value of threads is limited. But it should give a little extra speed boost. (You can see this for yourself in my inference kernel -- this is why it includes a speed test section.)
", "rawMarkdown": "I think you get only 2 CPU cores when GPU is enabled, so the value of threads is limited. But it should give a little extra speed boost. (You can see this for yourself in my inference kernel -- this is why it includes a speed test section.)", "votes": null } ], "comments": [ { "id": 752195, "author_name": "humananalog", "author_url": "", "post_date": "02/20/2020 19:34:52", "content": "Some things you can try:
\n\nHi, I see you using multi-threading to speed up the kernel, is this effective？\nHow do I know how many threads should I set？\nThank you
", "votes": null, "replies": [] }, { "id": 752441, "author_name": "akashnandi", "author_url": "", "post_date": "02/21/2020 02:51:37", "content": "You can query the number of CPUs available and then use that value for splitting work. But it's not really useful for this competition.
", "votes": null, "replies": [] }, { "id": 752768, "author_name": "humananalog", "author_url": "", "post_date": "02/21/2020 11:41:54", "content": "I think you get only 2 CPU cores when GPU is enabled, so the value of threads is limited. But it should give a little extra speed boost. (You can see this for yourself in my inference kernel -- this is why it includes a speed test section.)
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "752136": "On committing my notebook took around 3.5 hours. But on submitting, notebook timeout occurs. I searched and got to know on submitting our model would be run through 4000 videos (is it including 400 sample test videos or else 4000 + 400 videos?). As 9 hours is the limit, so 8.1 seconds per video for preprocessing + prediction should be the worst case. But my preprocessing itself is exceeding the time limit in which I am just simply capturing 40 frames per video and directly sending it to my model. How can one reduce the capture time?", "752195": "Some things you can try:\n\n- Use the GPU.\n- Do less: use smaller models, use fewer frames.\n- Use faster face extraction.\n- Don't read video frames you're not using.\n- Batch images together to use the GPU more effectively.", "752385": "Hi, I see you using multi-threading to speed up the kernel, is this effective？\nHow do I know how many threads should I set？\nThank you", "752441": "You can query the number of CPUs available and then use that value for splitting work. But it's not really useful for this competition.", "752768": "I think you get only 2 CPU cores when GPU is enabled, so the value of threads is limited. But it should give a little extra speed boost. (You can see this for yourself in my inference kernel -- this is why it includes a speed test section.)" }, "source": "meta" }