![\"image\"](\"https://user-images.githubusercontent.com/17668390/137544399-7652d99c-954b-4106-b997-6acb0c071127.png\")
A small list of some promising models in such given task.
\nModel: SOLOv2: Dynamic and Fast Instance Segmentation
\nCode: AdelaiDet/SOLOv2 - SOLOv1
Model: CenterMask : Real-Time Anchor-Free Instance Segmentation\n
\nCode: CenterMask2
![\"image\"](\"https://user-images.githubusercontent.com/17668390/137544725-0b7538ca-3af6-4fc4-b1f6-f5a3f3040121.png\")
Model: Mask R-CNN for Object Detection and Segmentation
\nCode: Mask-RCNN/Usiigaci
![\"T98Gelectrotaxis-1\"](\"https://user-images.githubusercontent.com/17668390/137575540-7e274dff-2a3a-4e1f-8151-4000a5be25b6.gif\")
![\"project_usiigaci2\"](\"https://user-images.githubusercontent.com/17668390/137545627-6633e6b0-3486-4698-abe5-9e986ea7820c.gif\")
Model: You Only Look At CoefficienTs
\nCode: YOLACT
![\"\"](\"https://raw.githubusercontent.com/dbolya/yolact/master/data/yolact_example_0.png\")
Model: PolarMask: Single Shot Instance Segmentation with Polar Representation
\nCode: PolarMask
![\"\"](\"https://user-images.githubusercontent.com/17668390/137545507-7ed6d0e8-a384-479b-8908-4706939ee10d.png\")
If you like, check out this survey paper on image segmentation. It's quite helpful.
\nPaper: Image Segmentation Using Deep Learning:- A Survey
![\"image\"](\"https://user-images.githubusercontent.com/17668390/137546231-0bd5fdce-52bd-4bf0-9aa3-ca01e2f6bb7d.png\")
A small list of some promising models in such given task.
\nModel: SOLOv2: Dynamic and Fast Instance Segmentation
\nCode: AdelaiDet/SOLOv2 - SOLOv1
Model: CenterMask : Real-Time Anchor-Free Instance Segmentation\n
\nCode: CenterMask2
Model: Mask R-CNN for Object Detection and Segmentation
\nCode: Mask-RCNN/Usiigaci
Model: You Only Look At CoefficienTs
\nCode: YOLACT
Model: PolarMask: Single Shot Instance Segmentation with Polar Representation
\nCode: PolarMask
If you like, check out this survey paper on image segmentation. It's quite helpful.
\nPaper: Image Segmentation Using Deep Learning:- A Survey
Thank you for this.
\nIt reinforces the sad truth that PyTorch dominates Tensorflow in terms of use ability for this application.
\nThe only implementation I’m aware of for TF is MaskRCNN which is in TF1 or using the OD zoo which is hideous.
\n—-
\nThat being said, I’ll try and leverage the AutoML EfficientDet implementation and see if I can make something work though!!
\nThanks again for compiling this, if I find anything else I’ll comment here.
", "rawMarkdown": "Thank you for this.\n\nIt reinforces the sad truth that PyTorch dominates Tensorflow in terms of use ability for this application.\n\nThe only implementation I’m aware of for TF is MaskRCNN which is in TF1 or using the OD zoo which is hideous.\n\n—-\n\nThat being said, I’ll try and leverage the AutoML EfficientDet implementation and see if I can make something work though!!\n\nThanks again for compiling this, if I find anything else I’ll comment here.", "votes": null }, { "id": "1546802", "postDate": "10/16/2021 14:18:00", "content": "@dschettler8845
\nThere was a good reason to switch entirely from TensorFlow to PyTorch and that's why now we see this noticeable gap. But now many things are changed on TensorFlow's (v2) side but still, it will need time to adapt to such changes and trust among researchers.
IMHO, for this competition, I think there are two things a tf / keras participant can do. If he/she competes to win, he/she better choose PyTorch here (unfortunately no other option) and otherwise, he/she should take this as an opportunity to learn how to reimplement these official PyTorch implementations in TensorFlow 2 best practice. [ Unofficial: YOLACT-TF - YOLACT. ]
Sadly, the well-known matterport/MaskRCNN, stopped updating their code to support the latest version. However, if you like you can see this non-official support for mask-rcnn with some additional features: maskrcnn_tf2
", "rawMarkdown": "dschettler8845 \nThere was a good reason to switch entirely from TensorFlow to PyTorch and that's why now we see this noticeable gap. But now many things are changed on TensorFlow's (v2) side but still, it will need time to adapt to such changes and trust among researchers. \n\nIMHO, for this competition, I think there are two things a `tf / keras` participant can do. If he/she competes to win, he/she better choose PyTorch here (unfortunately no other option) and otherwise, he/she should take this as an opportunity to learn how to **reimplement** these official PyTorch implementations in TensorFlow 2 best practice. [ Unofficial: [YOLACT-TF](https://github.com/leohsuofnthu/Tensorflow-YOLACT) - [YOLACT](https://github.com/anshkumar/yolact). ]\n\nSadly, the well-known [matterport/MaskRCNN](https://github.com/matterport/Mask_RCNN), stopped updating their code to support the latest version. However, if you like you can see this non-official support for mask-rcnn with some additional features: [maskrcnn_tf2](https://github.com/alexander-pv/maskrcnn_tf2)", "votes": null }, { "id": "1546958", "postDate": "10/16/2021 15:39:30", "content": "Even though I love using keras/TF2, Its really hard for some people to use it since some of the basic classification models have been missing from their packages, not to mention other segmentation/OD models. I doubt that the situation is going to be changed in the short term cause it's been a long time since last time keras-application getting updates.
", "rawMarkdown": "Even though I love using keras/TF2, Its really hard for some people to use it since some of the basic classification models have been missing from their packages, not to mention other segmentation/OD models. I doubt that the situation is going to be changed in the short term cause it's been a long time since last time keras-application getting updates.", "votes": null }, { "id": "1547003", "postDate": "10/16/2021 16:19:40", "content": "100%… ideally we get a conglomerate of awesome people coming together to make something like TFA.keras.applications where we can host all that kinda stuff.
\nI would be down to participate for sure.
", "rawMarkdown": "100%... ideally we get a conglomerate of awesome people coming together to make something like TFA.keras.applications where we can host all that kinda stuff.\n\nI would be down to participate for sure.", "votes": null }, { "id": "1548800", "postDate": "10/18/2021 14:34:58", "content": "@dschettler8845 @gdoong
\nCheck this out.
Hi, I'm tring to use it for this competition and have successfully trained several models, but I just don't know how to build this project correctly in the submission project, when use python setup.py build develop as described in the README, I just change the path of the build dir because of the input dir in kaggle is read-only, and actually I succeed to build but failed to import when using import adet.**, it said the module does not exist!
\nAny help will be appreciated!
Any sample code for that?
", "rawMarkdown": "Any sample code for that?", "votes": null }, { "id": "1581005", "postDate": "11/13/2021 09:56:56", "content": "@dschettler8845
\nI'm looking forward to seeing you start using pytorch to become a notebookGM!
Someday 🤛.
\nI think I’ll probably try to start releasing two versions of my notebooks… one in Pytorch and one in TF… but I’m going to have to spend some time learning first!!
", "rawMarkdown": "Someday 🤛.\n\nI think I’ll probably try to start releasing two versions of my notebooks… one in Pytorch and one in TF… but I’m going to have to spend some time learning first!!", "votes": null } ], "comments": [ { "id": 1546706, "author_name": "dschettler8845", "author_url": "", "post_date": "10/16/2021 12:50:53", "content": "Thank you for this.
\nIt reinforces the sad truth that PyTorch dominates Tensorflow in terms of use ability for this application.
\nThe only implementation I’m aware of for TF is MaskRCNN which is in TF1 or using the OD zoo which is hideous.
\n—-
\nThat being said, I’ll try and leverage the AutoML EfficientDet implementation and see if I can make something work though!!
\nThanks again for compiling this, if I find anything else I’ll comment here.
", "votes": null, "replies": [ { "id": 1546802, "author_name": "ipythonx", "author_url": "", "post_date": "10/16/2021 14:18:00", "content": "@dschettler8845
\nThere was a good reason to switch entirely from TensorFlow to PyTorch and that's why now we see this noticeable gap. But now many things are changed on TensorFlow's (v2) side but still, it will need time to adapt to such changes and trust among researchers.
IMHO, for this competition, I think there are two things a tf / keras participant can do. If he/she competes to win, he/she better choose PyTorch here (unfortunately no other option) and otherwise, he/she should take this as an opportunity to learn how to reimplement these official PyTorch implementations in TensorFlow 2 best practice. [ Unofficial: YOLACT-TF - YOLACT. ]
Sadly, the well-known matterport/MaskRCNN, stopped updating their code to support the latest version. However, if you like you can see this non-official support for mask-rcnn with some additional features: maskrcnn_tf2
", "votes": null, "replies": [] }, { "id": 1546958, "author_name": "gdoong", "author_url": "", "post_date": "10/16/2021 15:39:30", "content": "Even though I love using keras/TF2, Its really hard for some people to use it since some of the basic classification models have been missing from their packages, not to mention other segmentation/OD models. I doubt that the situation is going to be changed in the short term cause it's been a long time since last time keras-application getting updates.
", "votes": null, "replies": [] }, { "id": 1547003, "author_name": "dschettler8845", "author_url": "", "post_date": "10/16/2021 16:19:40", "content": "100%… ideally we get a conglomerate of awesome people coming together to make something like TFA.keras.applications where we can host all that kinda stuff.
\nI would be down to participate for sure.
", "votes": null, "replies": [] }, { "id": 1548800, "author_name": "ipythonx", "author_url": "", "post_date": "10/18/2021 14:34:58", "content": "@dschettler8845 @gdoong
\nCheck this out.
@dschettler8845
\nI'm looking forward to seeing you start using pytorch to become a notebookGM!
Someday 🤛.
\nI think I’ll probably try to start releasing two versions of my notebooks… one in Pytorch and one in TF… but I’m going to have to spend some time learning first!!
", "votes": null, "replies": [] } ] }, { "id": 1563546, "author_name": "alvinzhy", "author_url": "", "post_date": "10/28/2021 11:56:38", "content": "Hi, I'm tring to use it for this competition and have successfully trained several models, but I just don't know how to build this project correctly in the submission project, when use python setup.py build develop as described in the README, I just change the path of the build dir because of the input dir in kaggle is read-only, and actually I succeed to build but failed to import when using import adet.**, it said the module does not exist!
\nAny help will be appreciated!
Any sample code for that?
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1546104": "A small list of some promising models in such given task. \n\n**Model**: [SOLOv2: Dynamic and Fast Instance Segmentation](https://arxiv.org/pdf/2003.10152.pdf)\n**Code**: [AdelaiDet/SOLOv2 ](https://github.com/aim-uofa/AdelaiDet/blob/master/configs/SOLOv2/README.md) - [SOLOv1](https://github.com/WXinlong/SOLO)\n\n\n\n---\n\n**Model**: [CenterMask : Real-Time Anchor-Free Instance Segmentation\n](https://arxiv.org/pdf/1911.06667.pdf)\n**Code**: [CenterMask2](https://github.com/youngwanLEE/centermask2)\n\n\n\n---\n\n**Model**: [Mask R-CNN for Object Detection and Segmentation](https://arxiv.org/pdf/1703.06870.pdf)\n**Code**: [Mask-RCNN/Usiigaci](https://github.com/oist/usiigaci)\n\n \n\n---\n\n**Model**: [You Only Look At CoefficienTs](https://arxiv.org/pdf/1912.06218.pdf)\n**Code**: [YOLACT](https://github.com/dbolya/yolact)\n\n\n\n---\n\n**Model**: [PolarMask: Single Shot Instance Segmentation with Polar Representation](https://arxiv.org/pdf/1909.13226.pdf)\n**Code**: [PolarMask](https://github.com/xieenze/PolarMask)\n\n\n\n---\n\nIf you like, check out this survey paper on image segmentation. It's quite helpful.\nPaper: [Image Segmentation Using Deep Learning:- A Survey](https://arxiv.org/pdf/2001.05566.pdf)\n\n", "1546706": "Thank you for this.\n\nIt reinforces the sad truth that PyTorch dominates Tensorflow in terms of use ability for this application.\n\nThe only implementation I’m aware of for TF is MaskRCNN which is in TF1 or using the OD zoo which is hideous.\n\n—-\n\nThat being said, I’ll try and leverage the AutoML EfficientDet implementation and see if I can make something work though!!\n\nThanks again for compiling this, if I find anything else I’ll comment here.", "1546802": "dschettler8845 \nThere was a good reason to switch entirely from TensorFlow to PyTorch and that's why now we see this noticeable gap. But now many things are changed on TensorFlow's (v2) side but still, it will need time to adapt to such changes and trust among researchers. \n\nIMHO, for this competition, I think there are two things a `tf / keras` participant can do. If he/she competes to win, he/she better choose PyTorch here (unfortunately no other option) and otherwise, he/she should take this as an opportunity to learn how to **reimplement** these official PyTorch implementations in TensorFlow 2 best practice. [ Unofficial: [YOLACT-TF](https://github.com/leohsuofnthu/Tensorflow-YOLACT) - [YOLACT](https://github.com/anshkumar/yolact). ]\n\nSadly, the well-known [matterport/MaskRCNN](https://github.com/matterport/Mask_RCNN), stopped updating their code to support the latest version. However, if you like you can see this non-official support for mask-rcnn with some additional features: [maskrcnn_tf2](https://github.com/alexander-pv/maskrcnn_tf2)", "1546958": "Even though I love using keras/TF2, Its really hard for some people to use it since some of the basic classification models have been missing from their packages, not to mention other segmentation/OD models. I doubt that the situation is going to be changed in the short term cause it's been a long time since last time keras-application getting updates.", "1547003": "100%... ideally we get a conglomerate of awesome people coming together to make something like TFA.keras.applications where we can host all that kinda stuff.\n\nI would be down to participate for sure.", "1548800": "dschettler8845 @gdoong \nCheck [this](https://github.com/keras-team/governance/blob/master/rfcs/20200827-keras-cv-scoping-design.md#design-proposal) out.", "1563546": "Hi, I'm tring to use it for this competition and have successfully trained several models, but I just don't know how to build this project correctly in the submission project, when use `python setup.py build develop` as described in the README, I just change the path of the build dir because of the input dir in kaggle is read-only, and actually I succeed to build but failed to import when using `import adet.**`, it said the module does not exist!\nAny help will be appreciated!", "1566073": "Any sample code for that?", "1581005": "dschettler8845 \nI'm looking forward to seeing you start using pytorch to become a notebookGM!", "1581488": "Someday 🤛.\n\nI think I’ll probably try to start releasing two versions of my notebooks… one in Pytorch and one in TF… but I’m going to have to spend some time learning first!!" }, "source": "meta" }