![\"K1\"](\"https://i.ibb.co/kS9wQvs/LoFTR0.jpg\")
I start with images … since images say more than words …
\n
![\"K2\"](\"https://i.ibb.co/86mGkrX/LoFTR1.jpg\")
![\"K3\"](\"https://i.ibb.co/5hTh4Qm/LoFTR2.jpg\")
Second part with my NN automatic background removal pipeline described here DATASET - dorsal fins for all IDs without background:
\n![\"K7\"](\"https://i.ibb.co/yQzhzb8/Lo-FTR-BR1.jpg\")
![\"K8\"](\"https://i.ibb.co/HncZPKW/Lo-FTR-BR2.jpg\")
I decided to show you my research conducted in this competition connected with feature matching. This topic is implemented together with background removal.
\nIn notebook Whales feature matching LoFTR - Kornia I presented three things:
\nNotebook: https://www.kaggle.com/remekkinas/whales-feature-matching-loftr-kornia
", "messages": [ { "id": 1706462, "postDate": "2022-02-27T13:58:02.830Z", "content": "I start with images … since images say more than words …
\n
Second part with my NN automatic background removal pipeline described here DATASET - dorsal fins for all IDs without background:
\n
I decided to show you my research conducted in this competition connected with feature matching. This topic is implemented together with background removal.
\nIn notebook Whales feature matching LoFTR - Kornia I presented three things:
\nNotebook: https://www.kaggle.com/remekkinas/whales-feature-matching-loftr-kornia
", "rawMarkdown": "I start with images ... since images say more than words ...\n\n\n\n\n\n\n\nSecond part with my NN automatic background removal pipeline described here [DATASET - dorsal fins for all IDs without background](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/309214):\n\n\n\n\n\nI decided to show you my research conducted in this competition connected with feature matching. This topic is implemented together with background removal. \n\nIn notebook Whales feature matching LoFTR - Kornia I presented three things:\n\n- LoFTR - Detector-Free Local Feature Matching with Transformers\n- Kornia - absolutely fantastic library for Pytorch - Kornia is a differentiable library that allows classical computer vision to be integrated into deep learning models.\n- Feature matching for images with background and without.\n\nNotebook: https://www.kaggle.com/remekkinas/whales-feature-matching-loftr-kornia", "votes": 29 }, { "id": 1716686, "postDate": "2022-03-09T09:39:31.653Z", "content": "Thank you for sharing!!
\nLove your works!
Great work.
", "rawMarkdown": "Great work.", "votes": 1, "replies": [ { "id": 1713028, "postDate": "2022-03-05T15:43:41.303Z", "content": "Thank you very much! The more I am working with this library the more I am sure it will help me :) Looking now for tricky way to use LoFTR to improve score (but a little bit later - now working on new dataset).
", "rawMarkdown": "Thank you very much! The more I am working with this library the more I am sure it will help me :) Looking now for tricky way to use LoFTR to improve score (but a little bit later - now working on new dataset)." } ] }, { "id": 1710427, "postDate": "2022-03-03T03:15:49.760Z", "content": "@remekkinas great analogy of using kornia for this comp, even I have been using this library for my personal projects for a while now, could you share some light on how are you tackling the issue of ocean background removal
", "rawMarkdown": "@remekkinas great analogy of using kornia for this comp, even I have been using this library for my personal projects for a while now, could you share some light on how are you tackling the issue of ocean background removal", "votes": 1, "replies": [ { "id": 1713029, "postDate": "2022-03-05T15:44:39.223Z", "content": "Than you for your comment and feedback.
\nBackground removal notebook will be publishet tomorrow. Now finishing working on new dataset.
Thanks for sharing kornia !
\nIs the dataset in your notebook the cropped FIN?
Yes. I used notebook created by @phalanx . My is not published yet (still working on dataset - it requires more steps on cleaning data - however first models train quite good (without background removal) - CV ~0.79 ).
", "rawMarkdown": "Yes. I used notebook created by @phalanx . My is not published yet (still working on dataset - it requires more steps on cleaning data - however first models train quite good (without background removal) - CV ~0.79 ).", "votes": 1 }, { "id": 1706619, "postDate": "2022-02-27T16:55:36.123Z", "content": "@dragonzhang now it is updated - it shows how to filter information from background using pipeline I work on.
", "rawMarkdown": "@dragonzhang now it is updated - it shows how to filter information from background using pipeline I work on." } ] }, { "id": 1716625, "postDate": "2022-03-09T08:13:26.487Z", "content": "Thank you for sharing!
\nI have a feeling you're about to win a gold medal in this competition ;)
Hi @haqishen thank you! Some day … some day … This is my dream. You know what it is …. gold on LB and then … anything can happen :) But I enjoy progress and sharing. Trying to reach my second GM in Notebooks :) and then completly switch to competition.
\nBTW: you have great score too! Cool to see you in this competition.
", "rawMarkdown": "Hi @haqishen thank you! Some day ... some day ... This is my dream. You know what it is .... gold on LB and then ... anything can happen :) But I enjoy progress and sharing. Trying to reach my second GM in Notebooks :) and then completly switch to competition. \n\nBTW: you have great score too! Cool to see you in this competition.", "votes": 1 } ] }, { "id": 1713688, "postDate": "2022-03-06T09:07:14.617Z", "content": "Today I did SURF and LoFTR comparision …. Two observations:
\nSee comparison:
\n![\"Lo\"](\"https://i.ibb.co/smrJdDf/loftr01.jpg\")
\n![\"su\"](\"https://i.ibb.co/VQ7qhGr/surf01.jpg\")
Thank you! I am just coding for … fun. Love doing that :)
", "rawMarkdown": "Thank you! I am just coding for ... fun. Love doing that :)" } ] } ], "comments": [ { "id": 1716686, "author_name": "Haru", "author_url": "", "post_date": "2022-03-09T09:39:31.653000", "content": "Thank you for sharing!!
\nLove your works!
Great work.
", "votes": 1, "replies": [ { "id": 1713028, "author_name": "Remek Kinas", "author_url": "", "post_date": "2022-03-05T15:43:41.303000", "content": "Thank you very much! The more I am working with this library the more I am sure it will help me :) Looking now for tricky way to use LoFTR to improve score (but a little bit later - now working on new dataset).
", "votes": 0, "replies": [] } ] }, { "id": 1710427, "author_name": "Harsh", "author_url": "", "post_date": "2022-03-03T03:15:49.760000", "content": "@remekkinas great analogy of using kornia for this comp, even I have been using this library for my personal projects for a while now, could you share some light on how are you tackling the issue of ocean background removal
", "votes": 1, "replies": [ { "id": 1713029, "author_name": "Remek Kinas", "author_url": "", "post_date": "2022-03-05T15:44:39.223000", "content": "Than you for your comment and feedback.
\nBackground removal notebook will be publishet tomorrow. Now finishing working on new dataset.
Thanks for sharing kornia !
\nIs the dataset in your notebook the cropped FIN?
Yes. I used notebook created by @phalanx . My is not published yet (still working on dataset - it requires more steps on cleaning data - however first models train quite good (without background removal) - CV ~0.79 ).
", "votes": 1, "replies": [] }, { "id": 1706619, "author_name": "Remek Kinas", "author_url": "", "post_date": "2022-02-27T16:55:36.123000", "content": "@dragonzhang now it is updated - it shows how to filter information from background using pipeline I work on.
", "votes": 0, "replies": [] } ] }, { "id": 1716625, "author_name": "Qishen Ha", "author_url": "", "post_date": "2022-03-09T08:13:26.487000", "content": "Thank you for sharing!
\nI have a feeling you're about to win a gold medal in this competition ;)
Hi @haqishen thank you! Some day … some day … This is my dream. You know what it is …. gold on LB and then … anything can happen :) But I enjoy progress and sharing. Trying to reach my second GM in Notebooks :) and then completly switch to competition.
\nBTW: you have great score too! Cool to see you in this competition.
", "votes": 1, "replies": [] } ] }, { "id": 1713688, "author_name": "Remek Kinas", "author_url": "", "post_date": "2022-03-06T09:07:14.617000", "content": "Today I did SURF and LoFTR comparision …. Two observations:
\nSee comparison:
\n
\n
Thank you! I am just coding for … fun. Love doing that :)
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1706462": "I start with images ... since images say more than words ...\n\n\n\n\n\n\n\nSecond part with my NN automatic background removal pipeline described here [DATASET - dorsal fins for all IDs without background](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/309214):\n\n\n\n\n\nI decided to show you my research conducted in this competition connected with feature matching. This topic is implemented together with background removal. \n\nIn notebook Whales feature matching LoFTR - Kornia I presented three things:\n\n- LoFTR - Detector-Free Local Feature Matching with Transformers\n- Kornia - absolutely fantastic library for Pytorch - Kornia is a differentiable library that allows classical computer vision to be integrated into deep learning models.\n- Feature matching for images with background and without.\n\nNotebook: https://www.kaggle.com/remekkinas/whales-feature-matching-loftr-kornia", "1716686": "Thank you for sharing!!\nLove your works!", "1711845": "Great work.", "1710427": "@remekkinas great analogy of using kornia for this comp, even I have been using this library for my personal projects for a while now, could you share some light on how are you tackling the issue of ocean background removal", "1706467": "\n\nThanks for sharing kornia !\n\n\nIs the dataset in your notebook the cropped FIN?\n", "1716625": "Thank you for sharing!\nI have a feeling you're about to win a gold medal in this competition ;)", "1713688": "Today I did SURF and LoFTR comparision .... Two observations:\n- SURF and SIFT are not in Kaggle distribution - algorithm is patented and is excluded in this configuration \n- In my opinion SURF works worse then LoFTR on this dataset - it was difficult to find a match above 75% even on very similar fins (belonging to the same individual)\n\nSee comparison:\n\n ", "1717585": "", "1706736": "" } }