One month's work on yolo learning, model training, roughly trained 21 epochs ending up with no more credit card to take google cloud $300 free trial = .=\nIt would be so glad to hear you guys get much better score by using many more powerful GPUs, training with more epochs based on my solution.\nhttps://github.com/mingweihe/ImageNet
", "messages": [ { "id": "378099", "postDate": "08/30/2018 02:17:39", "content": "One month's work on yolo learning, model training, roughly trained 21 epochs ending up with no more credit card to take google cloud $300 free trial = .=\nIt would be so glad to hear you guys get much better score by using many more powerful GPUs, training with more epochs based on my solution.\nhttps://github.com/mingweihe/ImageNet
", "rawMarkdown": "One month's work on yolo learning, model training, roughly trained 21 epochs ending up with no more credit card to take google cloud $300 free trial = .=\nIt would be so glad to hear you guys get much better score by using many more powerful GPUs, training with more epochs based on my solution.\nhttps://github.com/mingweihe/ImageNet", "votes": null }, { "id": "428161", "postDate": "11/26/2018 21:30:03", "content": "Ming, thanks for showing us your code, it helped me understand how to put YOLO v3 into action.
", "rawMarkdown": "Ming, thanks for showing us your code, it helped me understand how to put YOLO v3 into action.", "votes": null }, { "id": "428695", "postDate": "11/27/2018 18:15:36", "content": "Dear Paul, you are welcome. happy to know that, good luck.
", "rawMarkdown": "Dear Paul, you are welcome. happy to know that, good luck.", "votes": null }, { "id": "436676", "postDate": "12/10/2018 18:19:07", "content": "Thanks for your Wonderful article too
", "rawMarkdown": "Thanks for your Wonderful article too", "votes": null }, { "id": "436710", "postDate": "12/10/2018 19:51:25", "content": "thank u for u'r code
", "rawMarkdown": "thank u for u'r code", "votes": null }, { "id": "436722", "postDate": "12/10/2018 20:31:11", "content": "You are quite welcome, thanks for reply.
", "rawMarkdown": "You are quite welcome, thanks for reply.", "votes": null }, { "id": "436724", "postDate": "12/10/2018 20:34:07", "content": "My pleasure. 👍
", "rawMarkdown": "My pleasure. 👍", "votes": null }, { "id": "465315", "postDate": "02/02/2019 19:49:20", "content": "I tried to download the dataset,but nothing is downloaded. username and password are fine. no error message is printed out.
", "rawMarkdown": "I tried to download the dataset,but nothing is downloaded. username and password are fine. no error message is printed out.", "votes": null }, { "id": "485774", "postDate": "03/07/2019 22:52:24", "content": "Hi, sorry for a late reply.\nIs that issue fixed?\nWould you like to share how you download the dataset.
", "rawMarkdown": "Hi, sorry for a late reply.\nIs that issue fixed?\nWould you like to share how you download the dataset.", "votes": null }, { "id": "506214", "postDate": "04/03/2019 06:11:57", "content": "Thanks for sharing the detailed steps.
\n\nI followed this tutorial, but had to make some changes, as my machine is running Ubuntu 16.04.6 with one RTX2080(8GB). So I had to go for Cuda-10 with Cudnn 7.4.2 (to support the Turing Architecture of the GPU) and OpenCV is 3.4.5. I also had to change the yolov3-ILSVRC.cfg training params. I changed the batch to 32 and subsamples to 8 so that I don't get the \"CUDA: Out of memory\" error.
\n\nTrained for 10000 iterations and my score was 0.509.
\n\nOne question I have in the setup is: Why didn't we set CUDNN=1 in the Makefile of darknet? Will it make any difference if we set that to 1?
", "rawMarkdown": "Thanks for sharing the detailed steps.\n\nI followed this tutorial, but had to make some changes, as my machine is running Ubuntu 16.04.6 with one RTX2080(8GB). So I had to go for Cuda-10 with Cudnn 7.4.2 (to support the Turing Architecture of the GPU) and OpenCV is 3.4.5. I also had to change the yolov3-ILSVRC.cfg training params. I changed the batch to 32 and subsamples to 8 so that I don't get the \"CUDA: Out of memory\" error. \n\nTrained for 10000 iterations and my score was 0.509. \n\nOne question I have in the setup is: Why didn't we set CUDNN=1 in the Makefile of darknet? Will it make any difference if we set that to 1?", "votes": null }, { "id": "508233", "postDate": "04/05/2019 20:31:39", "content": "Wow, that's awesome. Glad to hear you got this high score. That means my summary works.\nYour score is far more than my 0.98817, Haha.\nThanks for sharing it. \nAnd for Makefile part, I got some relevant errors when I set CUDNN to 1 last time. I haven't found a better solution yet.\nIf there's no error for your setup, just feel free to try it. I reckon the training processing would be faster.\nThank you again. :- )
", "rawMarkdown": "Wow, that's awesome. Glad to hear you got this high score. That means my summary works.\nYour score is far more than my 0.98817, Haha.\nThanks for sharing it. \nAnd for Makefile part, I got some relevant errors when I set CUDNN to 1 last time. I haven't found a better solution yet.\nIf there's no error for your setup, just feel free to try it. I reckon the training processing would be faster.\nThank you again. :- )", "votes": null }, { "id": "548894", "postDate": "06/10/2019 04:27:41", "content": "Great Work..!!\nBut i want to know that is there any other approach to solve this problem...?
", "rawMarkdown": "Great Work..!!\nBut i want to know that is there any other approach to solve this problem...?", "votes": null }, { "id": "735052", "postDate": "02/02/2020 12:57:26", "content": "Hello, where may the tutorial for imagenet be?
", "rawMarkdown": "Hello, where may the tutorial for imagenet be?", "votes": null } ], "comments": [ { "id": 428161, "author_name": "pdevine", "author_url": "", "post_date": "11/26/2018 21:30:03", "content": "Ming, thanks for showing us your code, it helped me understand how to put YOLO v3 into action.
", "votes": null, "replies": [ { "id": 428695, "author_name": "hemingwei", "author_url": "", "post_date": "11/27/2018 18:15:36", "content": "Dear Paul, you are welcome. happy to know that, good luck.
", "votes": null, "replies": [] }, { "id": 436676, "author_name": "kurianbenoy", "author_url": "", "post_date": "12/10/2018 18:19:07", "content": "Thanks for your Wonderful article too
", "votes": null, "replies": [] }, { "id": 436722, "author_name": "hemingwei", "author_url": "", "post_date": "12/10/2018 20:31:11", "content": "You are quite welcome, thanks for reply.
", "votes": null, "replies": [] } ] }, { "id": 436710, "author_name": "ayoubchebbi", "author_url": "", "post_date": "12/10/2018 19:51:25", "content": "thank u for u'r code
", "votes": null, "replies": [ { "id": 436724, "author_name": "hemingwei", "author_url": "", "post_date": "12/10/2018 20:34:07", "content": "My pleasure. 👍
", "votes": null, "replies": [] } ] }, { "id": 465315, "author_name": "tomsang", "author_url": "", "post_date": "02/02/2019 19:49:20", "content": "I tried to download the dataset,but nothing is downloaded. username and password are fine. no error message is printed out.
", "votes": null, "replies": [ { "id": 485774, "author_name": "hemingwei", "author_url": "", "post_date": "03/07/2019 22:52:24", "content": "Hi, sorry for a late reply.\nIs that issue fixed?\nWould you like to share how you download the dataset.
", "votes": null, "replies": [] } ] }, { "id": 506214, "author_name": "navaneethsen", "author_url": "", "post_date": "04/03/2019 06:11:57", "content": "Thanks for sharing the detailed steps.
\n\nI followed this tutorial, but had to make some changes, as my machine is running Ubuntu 16.04.6 with one RTX2080(8GB). So I had to go for Cuda-10 with Cudnn 7.4.2 (to support the Turing Architecture of the GPU) and OpenCV is 3.4.5. I also had to change the yolov3-ILSVRC.cfg training params. I changed the batch to 32 and subsamples to 8 so that I don't get the \"CUDA: Out of memory\" error.
\n\nTrained for 10000 iterations and my score was 0.509.
\n\nOne question I have in the setup is: Why didn't we set CUDNN=1 in the Makefile of darknet? Will it make any difference if we set that to 1?
", "votes": null, "replies": [ { "id": 508233, "author_name": "hemingwei", "author_url": "", "post_date": "04/05/2019 20:31:39", "content": "Wow, that's awesome. Glad to hear you got this high score. That means my summary works.\nYour score is far more than my 0.98817, Haha.\nThanks for sharing it. \nAnd for Makefile part, I got some relevant errors when I set CUDNN to 1 last time. I haven't found a better solution yet.\nIf there's no error for your setup, just feel free to try it. I reckon the training processing would be faster.\nThank you again. :- )
", "votes": null, "replies": [] }, { "id": 735052, "author_name": "hashtagkehl", "author_url": "", "post_date": "02/02/2020 12:57:26", "content": "Hello, where may the tutorial for imagenet be?
", "votes": null, "replies": [] } ] }, { "id": 548894, "author_name": "darshan206", "author_url": "", "post_date": "06/10/2019 04:27:41", "content": "Great Work..!!\nBut i want to know that is there any other approach to solve this problem...?
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "378099": "One month's work on yolo learning, model training, roughly trained 21 epochs ending up with no more credit card to take google cloud $300 free trial = .=\nIt would be so glad to hear you guys get much better score by using many more powerful GPUs, training with more epochs based on my solution.\nhttps://github.com/mingweihe/ImageNet", "428161": "Ming, thanks for showing us your code, it helped me understand how to put YOLO v3 into action.", "428695": "Dear Paul, you are welcome. happy to know that, good luck.", "436676": "Thanks for your Wonderful article too", "436710": "thank u for u'r code", "436722": "You are quite welcome, thanks for reply.", "436724": "My pleasure. 👍", "465315": "I tried to download the dataset,but nothing is downloaded. username and password are fine. no error message is printed out.", "485774": "Hi, sorry for a late reply.\nIs that issue fixed?\nWould you like to share how you download the dataset.", "506214": "Thanks for sharing the detailed steps.\n\nI followed this tutorial, but had to make some changes, as my machine is running Ubuntu 16.04.6 with one RTX2080(8GB). So I had to go for Cuda-10 with Cudnn 7.4.2 (to support the Turing Architecture of the GPU) and OpenCV is 3.4.5. I also had to change the yolov3-ILSVRC.cfg training params. I changed the batch to 32 and subsamples to 8 so that I don't get the \"CUDA: Out of memory\" error. \n\nTrained for 10000 iterations and my score was 0.509. \n\nOne question I have in the setup is: Why didn't we set CUDNN=1 in the Makefile of darknet? Will it make any difference if we set that to 1?", "508233": "Wow, that's awesome. Glad to hear you got this high score. That means my summary works.\nYour score is far more than my 0.98817, Haha.\nThanks for sharing it. \nAnd for Makefile part, I got some relevant errors when I set CUDNN to 1 last time. I haven't found a better solution yet.\nIf there's no error for your setup, just feel free to try it. I reckon the training processing would be faster.\nThank you again. :- )", "548894": "Great Work..!!\nBut i want to know that is there any other approach to solve this problem...?", "735052": "Hello, where may the tutorial for imagenet be?" }, "source": "meta" }