In the morning survey, I found MS R-CNN.
\nThis models was introduced in 2019 and outperformed Mask R-CNN.
\nI am not familiar with instance segmentation but we may improve the accuracy compared to the model based on Mask R-CNN.
paper
\nhttps://arxiv.org/abs/1903.00241
repository
\nhttps://github.com/zjhuang22/maskscoring_rcnn
slide share (Japanese)
\nhttps://www.slideshare.net/YusukeFujimoto1/paper-lt-mask-scoring-rcnn
I forked the repository.
\nAnd I fixed some errors according to this comments so you can run the demo notebook.
\nhttps://github.com/r-matsuzaka/maskscoring_rcnn
In the morning survey, I found MS R-CNN.
\nThis models was introduced in 2019 and outperformed Mask R-CNN.
\nI am not familiar with instance segmentation but we may improve the accuracy compared to the model based on Mask R-CNN.
paper
\nhttps://arxiv.org/abs/1903.00241
repository
\nhttps://github.com/zjhuang22/maskscoring_rcnn
slide share (Japanese)
\nhttps://www.slideshare.net/YusukeFujimoto1/paper-lt-mask-scoring-rcnn
I forked the repository.
\nAnd I fixed some errors according to this comments so you can run the demo notebook.
\nhttps://github.com/r-matsuzaka/maskscoring_rcnn
Also considered this one but didn't have time to test it out yet. Will be extremely grateful if you can share your findings afterwards :)
\nBtw, after skimming through this paper, I thought that a gradient-stopping operation should be added at the branching point to IoU head, so that the gradients propagated from that head won't be propagated further to the backbone. My intuition is that the gradients from this head will at some point \"stand in the way\" of the gradients from other heads. What do you think?
", "rawMarkdown": "Also considered this one but didn't have time to test it out yet. Will be extremely grateful if you can share your findings afterwards :)\n\nBtw, after skimming through this paper, I thought that a gradient-stopping operation should be added at the branching point to IoU head, so that the gradients propagated from that head won't be propagated further to the backbone. My intuition is that the gradients from this head will at some point \"stand in the way\" of the gradients from other heads. What do you think?", "votes": 1, "replies": [ { "id": 1589235, "postDate": "2021-11-20T04:22:54.767Z", "content": "Thank you very much for informative comments. I appreciate for it.Excuse me , actually I am reading this paper. I will finish reading this paper in this holiday. If I have any insight or thouts I will comment here.
\nThank you. Let's keep discuss it.
I add the slide share(Japanese).
\nhttps://www.slideshare.net/YusukeFujimoto1/paper-lt-mask-scoring-rcnn
This issue will help us to run demo notebook of MS-RCNN.
\nI will try it on my own PC and will prepare the notebook for the task.
Can you provide a jupyter notebook demo? #27
", "rawMarkdown": "This issue will help us to run demo notebook of MS-RCNN.\nI will try it on my own PC and will prepare the notebook for the task.\n\n[Can you provide a jupyter notebook demo? #27](https://github.com/zjhuang22/maskscoring_rcnn/issues/27#issuecomment-478257021)", "replies": [ { "id": 1591253, "postDate": "2021-11-22T05:56:36.217Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1591283, "postDate": "2021-11-22T06:29:07.853Z", "content": "I could managed to run the notebook.
", "rawMarkdown": "I could managed to run the notebook." } ] }, { "id": 1586458, "postDate": "2021-11-18T05:20:25.423Z", "content": "Thanks for sharing @osamurai :D
", "rawMarkdown": "Thanks for sharing @osamurai :D", "votes": 1 }, { "id": 1592769, "postDate": "2021-11-23T10:43:04.457Z", "content": "Thanks for sharing
", "rawMarkdown": "Thanks for sharing" } ], "comments": [ { "id": 1589155, "author_name": "Chan Kha Vu", "author_url": "", "post_date": "2021-11-20T01:48:55.750000", "content": "Also considered this one but didn't have time to test it out yet. Will be extremely grateful if you can share your findings afterwards :)
\nBtw, after skimming through this paper, I thought that a gradient-stopping operation should be added at the branching point to IoU head, so that the gradients propagated from that head won't be propagated further to the backbone. My intuition is that the gradients from this head will at some point \"stand in the way\" of the gradients from other heads. What do you think?
", "votes": 1, "replies": [ { "id": 1589235, "author_name": "John Doe", "author_url": "", "post_date": "2021-11-20T04:22:54.767000", "content": "Thank you very much for informative comments. I appreciate for it.Excuse me , actually I am reading this paper. I will finish reading this paper in this holiday. If I have any insight or thouts I will comment here.
\nThank you. Let's keep discuss it.
I add the slide share(Japanese).
\nhttps://www.slideshare.net/YusukeFujimoto1/paper-lt-mask-scoring-rcnn
This issue will help us to run demo notebook of MS-RCNN.
\nI will try it on my own PC and will prepare the notebook for the task.
Can you provide a jupyter notebook demo? #27
", "votes": 0, "replies": [ { "id": 1591253, "author_name": "", "author_url": "", "post_date": "2021-11-22T05:56:36.217000", "content": "", "votes": 0, "replies": [] }, { "id": 1591283, "author_name": "John Doe", "author_url": "", "post_date": "2021-11-22T06:29:07.853000", "content": "I could managed to run the notebook.
", "votes": 0, "replies": [] } ] }, { "id": 1586458, "author_name": "somuSan", "author_url": "", "post_date": "2021-11-18T05:20:25.423000", "content": "Thanks for sharing @osamurai :D
", "votes": 1, "replies": [] }, { "id": 1592769, "author_name": "Yunjie Li", "author_url": "", "post_date": "2021-11-23T10:43:04.457000", "content": "Thanks for sharing
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1586293": "In the morning survey, I found MS R-CNN.\nThis models was introduced in 2019 and outperformed Mask R-CNN.\nI am not familiar with instance segmentation but we may improve the accuracy compared to the model based on Mask R-CNN.\n\npaper\nhttps://arxiv.org/abs/1903.00241\n\nrepository\nhttps://github.com/zjhuang22/maskscoring_rcnn\n\nslide share (Japanese)\nhttps://www.slideshare.net/YusukeFujimoto1/paper-lt-mask-scoring-rcnn\n\nI forked the repository. \nAnd I fixed some errors according to [this comments](https://github.com/zjhuang22/maskscoring_rcnn/issues/27#issuecomment-478257021) so you can run the demo notebook.\nhttps://github.com/r-matsuzaka/maskscoring_rcnn", "1589155": "Also considered this one but didn't have time to test it out yet. Will be extremely grateful if you can share your findings afterwards :)\n\nBtw, after skimming through this paper, I thought that a gradient-stopping operation should be added at the branching point to IoU head, so that the gradients propagated from that head won't be propagated further to the backbone. My intuition is that the gradients from this head will at some point \"stand in the way\" of the gradients from other heads. What do you think?", "1587839": "I add the slide share(Japanese).\nhttps://www.slideshare.net/YusukeFujimoto1/paper-lt-mask-scoring-rcnn", "1590489": "This issue will help us to run demo notebook of MS-RCNN.\nI will try it on my own PC and will prepare the notebook for the task.\n\n[Can you provide a jupyter notebook demo? #27](https://github.com/zjhuang22/maskscoring_rcnn/issues/27#issuecomment-478257021)", "1586458": "Thanks for sharing @osamurai :D", "1592769": "Thanks for sharing" } }