We are using pytorch and we got some GPU resources. We currently have a classifier with 98.2% accuracy for empty masks, a variety of models (unet+res34 and some models created in TGS competition) with best iou of 0.8+ to detect ships, and some post-processing methods on masks. Have not performed the model ensemble yet.
\n\nFeels like we need some fresh thoughts right now. Please let me know if you are interested in joining us.
", "messages": [ { "id": 416675, "postDate": "2018-11-07T04:26:32.040Z", "content": "We are using pytorch and we got some GPU resources. We currently have a classifier with 98.2% accuracy for empty masks, a variety of models (unet+res34 and some models created in TGS competition) with best iou of 0.8+ to detect ships, and some post-processing methods on masks. Have not performed the model ensemble yet.
\n\nFeels like we need some fresh thoughts right now. Please let me know if you are interested in joining us.
", "rawMarkdown": "We are using pytorch and we got some GPU resources. We currently have a classifier with 98.2% accuracy for empty masks, a variety of models (unet+res34 and some models created in TGS competition) with best iou of 0.8+ to detect ships, and some post-processing methods on masks. Have not performed the model ensemble yet. \n\nFeels like we need some fresh thoughts right now. Please let me know if you are interested in joining us.", "votes": 2 }, { "id": 419209, "postDate": "2018-11-11T14:31:29.360Z", "content": "Can i team up with you? I have a GTX1080 TI and a GTX1060 on two differents computers.
", "rawMarkdown": "Can i team up with you? I have a GTX1080 TI and a GTX1060 on two differents computers.", "replies": [ { "id": 419437, "postDate": "2018-11-12T00:46:28.620Z", "content": "The merge deadline alreday passed
", "rawMarkdown": "The merge deadline alreday passed" } ] }, { "id": 416952, "postDate": "2018-11-07T13:59:21.767Z", "content": "I haven't got the gpu resources, due to that I am stuck at stage 2 with an IoU of 0.80 and dice 0.898. With more epochs I am sure it'll improve, but couldn't move forward to stage 3. I have the implementation but less gpu, hope you'd consider me for a team up. I am using pytorch and fastai. I am sending you a merge request.
", "rawMarkdown": "I haven't got the gpu resources, due to that I am stuck at stage 2 with an IoU of 0.80 and dice 0.898. With more epochs I am sure it'll improve, but couldn't move forward to stage 3. I have the implementation but less gpu, hope you'd consider me for a team up. I am using pytorch and fastai. I am sending you a merge request.", "replies": [ { "id": 416986, "postDate": "2018-11-07T15:11:19.133Z", "content": "wondering what is your current model?
", "rawMarkdown": "wondering what is your current model?", "votes": 1 }, { "id": 416990, "postDate": "2018-11-07T15:14:34.353Z", "content": "It is a Unet model with resnet34 as encoder. I have tried resnet50 too but no improvement. Moreover, resnet34 is faster in comparison to resnet50. I have added few modules to my architecture, which gave better improvements.
", "rawMarkdown": "It is a Unet model with resnet34 as encoder. I have tried resnet50 too but no improvement. Moreover, resnet34 is faster in comparison to resnet50. I have added few modules to my architecture, which gave better improvements." }, { "id": 417125, "postDate": "2018-11-07T19:55:38.707Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 416899, "postDate": "2018-11-07T12:17:49.903Z", "content": "Can I team up with you ?\nI am using caffe2 and pytorch (fastai). I only have one GTX1060 GPU. I used Detectron's Faster RCNN model to get bounding boxes. Then I used a unet+senet50 model to perform segmentations on bounding boxes with high cls score. My Faster RCNN model is still on training and my unet+senet50 model has 0.9+ iou.
", "rawMarkdown": "Can I team up with you ?\nI am using caffe2 and pytorch (fastai). I only have one GTX1060 GPU. I used Detectron's Faster RCNN model to get bounding boxes. Then I used a unet+senet50 model to perform segmentations on bounding boxes with high cls score. My Faster RCNN model is still on training and my unet+senet50 model has 0.9+ iou. ", "replies": [ { "id": 416995, "postDate": "2018-11-07T15:30:31.200Z", "content": "Hello, we have accepted the request. Please check your inbox and contact us asap.
", "rawMarkdown": "Hello, we have accepted the request. Please check your inbox and contact us asap." }, { "id": 417026, "postDate": "2018-11-07T16:24:51.860Z", "content": "Thanks for teaming up with me! Here is my email, zhx1586@126.com.
", "rawMarkdown": "Thanks for teaming up with me! Here is my email, zhx1586@126.com." }, { "id": 417035, "postDate": "2018-11-07T16:32:30.873Z", "rawMarkdown": "", "isDeleted": true } ] } ], "comments": [ { "id": 419209, "author_name": "Pierre Tisseur", "author_url": "", "post_date": "2018-11-11T14:31:29.360000", "content": "Can i team up with you? I have a GTX1080 TI and a GTX1060 on two differents computers.
", "votes": 0, "replies": [ { "id": 419437, "author_name": "Strideradu", "author_url": "", "post_date": "2018-11-12T00:46:28.620000", "content": "The merge deadline alreday passed
", "votes": 0, "replies": [] } ] }, { "id": 416952, "author_name": "PiGram", "author_url": "", "post_date": "2018-11-07T13:59:21.767000", "content": "I haven't got the gpu resources, due to that I am stuck at stage 2 with an IoU of 0.80 and dice 0.898. With more epochs I am sure it'll improve, but couldn't move forward to stage 3. I have the implementation but less gpu, hope you'd consider me for a team up. I am using pytorch and fastai. I am sending you a merge request.
", "votes": 0, "replies": [ { "id": 416986, "author_name": "Strideradu", "author_url": "", "post_date": "2018-11-07T15:11:19.133000", "content": "wondering what is your current model?
", "votes": 1, "replies": [] }, { "id": 416990, "author_name": "PiGram", "author_url": "", "post_date": "2018-11-07T15:14:34.353000", "content": "It is a Unet model with resnet34 as encoder. I have tried resnet50 too but no improvement. Moreover, resnet34 is faster in comparison to resnet50. I have added few modules to my architecture, which gave better improvements.
", "votes": 0, "replies": [] }, { "id": 417125, "author_name": "", "author_url": "", "post_date": "2018-11-07T19:55:38.707000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 416899, "author_name": "zhx1586", "author_url": "", "post_date": "2018-11-07T12:17:49.903000", "content": "Can I team up with you ?\nI am using caffe2 and pytorch (fastai). I only have one GTX1060 GPU. I used Detectron's Faster RCNN model to get bounding boxes. Then I used a unet+senet50 model to perform segmentations on bounding boxes with high cls score. My Faster RCNN model is still on training and my unet+senet50 model has 0.9+ iou.
", "votes": 0, "replies": [ { "id": 416995, "author_name": "Daniel", "author_url": "", "post_date": "2018-11-07T15:30:31.200000", "content": "Hello, we have accepted the request. Please check your inbox and contact us asap.
", "votes": 0, "replies": [] }, { "id": 417026, "author_name": "zhx1586", "author_url": "", "post_date": "2018-11-07T16:24:51.860000", "content": "Thanks for teaming up with me! Here is my email, zhx1586@126.com.
", "votes": 0, "replies": [] }, { "id": 417035, "author_name": "", "author_url": "", "post_date": "2018-11-07T16:32:30.873000", "content": "", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "416675": "We are using pytorch and we got some GPU resources. We currently have a classifier with 98.2% accuracy for empty masks, a variety of models (unet+res34 and some models created in TGS competition) with best iou of 0.8+ to detect ships, and some post-processing methods on masks. Have not performed the model ensemble yet. \n\nFeels like we need some fresh thoughts right now. Please let me know if you are interested in joining us.", "419209": "Can i team up with you? I have a GTX1080 TI and a GTX1060 on two differents computers.", "416952": "I haven't got the gpu resources, due to that I am stuck at stage 2 with an IoU of 0.80 and dice 0.898. With more epochs I am sure it'll improve, but couldn't move forward to stage 3. I have the implementation but less gpu, hope you'd consider me for a team up. I am using pytorch and fastai. I am sending you a merge request.", "416899": "Can I team up with you ?\nI am using caffe2 and pytorch (fastai). I only have one GTX1060 GPU. I used Detectron's Faster RCNN model to get bounding boxes. Then I used a unet+senet50 model to perform segmentations on bounding boxes with high cls score. My Faster RCNN model is still on training and my unet+senet50 model has 0.9+ iou. " } }