Does anyone face a problem that prediction on test set taking too long?\nIn my case, it takes around 10 hours. I'm using U-net on Keras.
", "messages": [ { "id": 372318, "postDate": "2018-08-19T00:02:15.307Z", "content": "Does anyone face a problem that prediction on test set taking too long?\nIn my case, it takes around 10 hours. I'm using U-net on Keras.
", "rawMarkdown": "Does anyone face a problem that prediction on test set taking too long?\nIn my case, it takes around 10 hours. I'm using U-net on Keras.", "votes": 3 }, { "id": 372654, "postDate": "2018-08-20T02:00:15.297Z", "content": "https://www.kaggle.com/hmendonca/u-net-model-with-submission\nthis kernel contains the fastest code I came up so far, hope it helps, but let me know if you have a better alternative ;) cheers
", "rawMarkdown": "https://www.kaggle.com/hmendonca/u-net-model-with-submission\nthis kernel contains the fastest code I came up so far, hope it helps, but let me know if you have a better alternative ;) cheers", "votes": 2 }, { "id": 374868, "postDate": "2018-08-24T03:09:13.883Z", "content": "For me it's about an hour on P100 with Pytorch using Unet34. But with 8 fold TTA it's going to take almost 10 hours( Do you also use TTA? it may be the reason.\nAlso regarding Keras, which back-end do you use, Tensorflow or Theano? Tensorflow gives almost the same performance as Pytorch, but Theano is quite sluggish.
", "rawMarkdown": "For me it's about an hour on P100 with Pytorch using Unet34. But with 8 fold TTA it's going to take almost 10 hours( Do you also use TTA? it may be the reason.\nAlso regarding Keras, which back-end do you use, Tensorflow or Theano? Tensorflow gives almost the same performance as Pytorch, but Theano is quite sluggish.", "replies": [ { "id": 375491, "postDate": "2018-08-25T09:52:43.963Z", "content": "It's TTA and Tensorflow backend. Thank you for your information.
", "rawMarkdown": "It's TTA and Tensorflow backend. Thank you for your information." } ] }, { "id": 372533, "postDate": "2018-08-19T16:10:45.627Z", "content": "I need 1.5h on an i5 with a GTX 1070
", "rawMarkdown": "I need 1.5h on an i5 with a GTX 1070", "replies": [ { "id": 372611, "postDate": "2018-08-19T21:33:10.713Z", "content": "Hello Alexis,\nWhich framework are you using (keras, tensorflow, pytorch, others)?\nare you using the UNET as well?\nThank you.
", "rawMarkdown": "Hello Alexis,\nWhich framework are you using (keras, tensorflow, pytorch, others)?\nare you using the UNET as well?\nThank you." } ] }, { "id": 375516, "postDate": "2018-08-25T11:01:16.287Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 375533, "postDate": "2018-08-25T11:45:01.743Z", "content": "If you need your computer for other tasks it might be difficult indeed but not impossible.
", "rawMarkdown": "If you need your computer for other tasks it might be difficult indeed but not impossible.", "votes": 1 }, { "id": 375541, "postDate": "2018-08-25T12:17:00.047Z", "rawMarkdown": "", "isDeleted": true }, { "id": 375548, "postDate": "2018-08-25T12:33:19.763Z", "content": "Hmm, with 4GB mem I think it's not quite enough for this competition. I have 1080ti. I think you could try to resize images smaller (e.g: ~200 x 200).
", "rawMarkdown": "Hmm, with 4GB mem I think it's not quite enough for this competition. I have 1080ti. I think you could try to resize images smaller (e.g: ~200 x 200).", "votes": 1 }, { "id": 375562, "postDate": "2018-08-25T13:33:30.117Z", "rawMarkdown": "", "isDeleted": true }, { "id": 376650, "postDate": "2018-08-27T22:17:54.030Z", "content": "you always can train on smaller crops instead of resizing
", "rawMarkdown": "you always can train on smaller crops instead of resizing" } ] } ], "comments": [ { "id": 372654, "author_name": "Henrique Mendonça", "author_url": "", "post_date": "2018-08-20T02:00:15.297000", "content": "https://www.kaggle.com/hmendonca/u-net-model-with-submission\nthis kernel contains the fastest code I came up so far, hope it helps, but let me know if you have a better alternative ;) cheers
", "votes": 2, "replies": [] }, { "id": 374868, "author_name": "Iafoss", "author_url": "", "post_date": "2018-08-24T03:09:13.883000", "content": "For me it's about an hour on P100 with Pytorch using Unet34. But with 8 fold TTA it's going to take almost 10 hours( Do you also use TTA? it may be the reason.\nAlso regarding Keras, which back-end do you use, Tensorflow or Theano? Tensorflow gives almost the same performance as Pytorch, but Theano is quite sluggish.
", "votes": 0, "replies": [ { "id": 375491, "author_name": "GO FOR IT", "author_url": "", "post_date": "2018-08-25T09:52:43.963000", "content": "It's TTA and Tensorflow backend. Thank you for your information.
", "votes": 0, "replies": [] } ] }, { "id": 372533, "author_name": "Alexis Letulier", "author_url": "", "post_date": "2018-08-19T16:10:45.627000", "content": "I need 1.5h on an i5 with a GTX 1070
", "votes": 0, "replies": [ { "id": 372611, "author_name": "GO FOR IT", "author_url": "", "post_date": "2018-08-19T21:33:10.713000", "content": "Hello Alexis,\nWhich framework are you using (keras, tensorflow, pytorch, others)?\nare you using the UNET as well?\nThank you.
", "votes": 0, "replies": [] } ] }, { "id": 375516, "author_name": "", "author_url": "", "post_date": "2018-08-25T11:01:16.287000", "content": "", "votes": 0, "replies": [ { "id": 375533, "author_name": "Alexis Letulier", "author_url": "", "post_date": "2018-08-25T11:45:01.743000", "content": "If you need your computer for other tasks it might be difficult indeed but not impossible.
", "votes": 1, "replies": [] }, { "id": 375541, "author_name": "", "author_url": "", "post_date": "2018-08-25T12:17:00.047000", "content": "", "votes": 0, "replies": [] }, { "id": 375548, "author_name": "GO FOR IT", "author_url": "", "post_date": "2018-08-25T12:33:19.763000", "content": "Hmm, with 4GB mem I think it's not quite enough for this competition. I have 1080ti. I think you could try to resize images smaller (e.g: ~200 x 200).
", "votes": 1, "replies": [] }, { "id": 375562, "author_name": "", "author_url": "", "post_date": "2018-08-25T13:33:30.117000", "content": "", "votes": 0, "replies": [] }, { "id": 376650, "author_name": "Emil", "author_url": "", "post_date": "2018-08-27T22:17:54.030000", "content": "you always can train on smaller crops instead of resizing
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "372318": "Does anyone face a problem that prediction on test set taking too long?\nIn my case, it takes around 10 hours. I'm using U-net on Keras.", "372654": "https://www.kaggle.com/hmendonca/u-net-model-with-submission\nthis kernel contains the fastest code I came up so far, hope it helps, but let me know if you have a better alternative ;) cheers", "374868": "For me it's about an hour on P100 with Pytorch using Unet34. But with 8 fold TTA it's going to take almost 10 hours( Do you also use TTA? it may be the reason.\nAlso regarding Keras, which back-end do you use, Tensorflow or Theano? Tensorflow gives almost the same performance as Pytorch, but Theano is quite sluggish.", "372533": "I need 1.5h on an i5 with a GTX 1070", "375516": "" } }