Hi All! \nThere were couple of very useful discussions on using RetinaNet implemented in Keras\nhttps://github.com/fizyr/keras-retinanet/\nI started late with it and could only run a few experiments with it. Some of my parameters and results for stage-1 are here:\nhttps://docs.google.com/spreadsheets/d/18ln0yxA19OZ4q6Pf780LD4r8BYhRKZ2xgZ3aQpcJ2uw/edit?usp=sharing
\n\nI recall some users had great results. @Shai you got 0.203 with a single retina-net model right ? \nIf you used RetinaNet, could you please add your parameters to this google doc ? It would help everyone learn! \nThanks for all the great discussions you all! :) \nMehul
", "messages": [ { "id": "414023", "postDate": "11/02/2018 01:35:51", "content": "Hi All! \nThere were couple of very useful discussions on using RetinaNet implemented in Keras\nhttps://github.com/fizyr/keras-retinanet/\nI started late with it and could only run a few experiments with it. Some of my parameters and results for stage-1 are here:\nhttps://docs.google.com/spreadsheets/d/18ln0yxA19OZ4q6Pf780LD4r8BYhRKZ2xgZ3aQpcJ2uw/edit?usp=sharing
\n\nI recall some users had great results. @Shai you got 0.203 with a single retina-net model right ? \nIf you used RetinaNet, could you please add your parameters to this google doc ? It would help everyone learn! \nThanks for all the great discussions you all! :) \nMehul
", "rawMarkdown": "Hi All! \nThere were couple of very useful discussions on using RetinaNet implemented in Keras\nhttps://github.com/fizyr/keras-retinanet/\nI started late with it and could only run a few experiments with it. Some of my parameters and results for stage-1 are here:\nhttps://docs.google.com/spreadsheets/d/18ln0yxA19OZ4q6Pf780LD4r8BYhRKZ2xgZ3aQpcJ2uw/edit?usp=sharing\n\nI recall some users had great results. @Shai you got 0.203 with a single retina-net model right ? \nIf you used RetinaNet, could you please add your parameters to this google doc ? It would help everyone learn! \nThanks for all the great discussions you all! :) \nMehul", "votes": null }, { "id": "414489", "postDate": "11/02/2018 21:25:44", "content": "Updated. Actually one of Branden's model lb was 0.203 in public lb. I did not check mine individually during stage 1. However, still some of my individual model scored around 0.20 in stage 2 private lb.
", "rawMarkdown": "Updated. Actually one of Branden's model lb was 0.203 in public lb. I did not check mine individually during stage 1. However, still some of my individual model scored around 0.20 in stage 2 private lb.", "votes": null }, { "id": "414492", "postDate": "11/02/2018 21:47:10", "content": "Awesome! Thanks Shai! \nOnce more people add the parameters I will make a git repo like this one:
\n\n\n\nI made this one for Mask-RCNN after DSB_2018.
", "rawMarkdown": "Awesome! Thanks Shai! \nOnce more people add the parameters I will make a git repo like this one: \n\nhttps://github.com/mpsampat/DSB_2018_Parameter_Selection_For_Matterport_Mask_RCNN/blob/master/comparison_of_Matterport_Mask_RCNN_parameters.md\n\nI made this one for Mask-RCNN after DSB_2018.", "votes": null }, { "id": "414498", "postDate": "11/02/2018 22:20:51", "content": "Thanks a lot for sharing your parameters! I just used a single RetinaNet for this competition and scored around 0.19 in private lb.
", "rawMarkdown": "Thanks a lot for sharing your parameters! I just used a single RetinaNet for this competition and scored around 0.19 in private lb.", "votes": null }, { "id": "414521", "postDate": "11/02/2018 23:10:20", "content": "Thanks for sharing guys, added mine too.
", "rawMarkdown": "Thanks for sharing guys, added mine too.", "votes": null }, { "id": "414535", "postDate": "11/02/2018 23:32:59", "content": "@Shai, @hxy777 and @taindow : thanks so much for filling out the google sheet :)
", "rawMarkdown": "Shai, @hxy777 and @taindow : thanks so much for filling out the google sheet :)", "votes": null }, { "id": "416063", "postDate": "11/06/2018 05:34:30", "content": "If i don't see this result, I certainly won't believe that ensemble has such a good effect. Thanks for sharing.
", "rawMarkdown": "If i don't see this result, I certainly won't believe that ensemble has such a good effect. Thanks for sharing.", "votes": null }, { "id": "416798", "postDate": "11/07/2018 09:54:44", "content": "Nice initiative, @Mehul. Our team is going to publish our solution later this week. I have started filling some parameters in your google doc, though.
", "rawMarkdown": "Nice initiative, @Mehul. Our team is going to publish our solution later this week. I have started filling some parameters in your google doc, though.", "votes": null }, { "id": "745133", "postDate": "02/13/2020 14:37:37", "content": "it's been a long time but i have 2 good resources.\nFirst, Stanford Drone Dataset from Priyanka Dwivedi\nfirst repo: https://github.com/priya-dwivedi/aerial_pedestrian_detection/blob/master/README.md\nsecond sharing and in Turkey there is a contest Teknofest and Sıddık Acıl team used RetinaNet image classification but this repo is Turkish, maybe benefit Turkish users\nsecond repo: https://github.com/huma-teknofest/Keras-RetinaNet-for-Teknofest-2019
", "rawMarkdown": "it's been a long time but i have 2 good resources.\nFirst, Stanford Drone Dataset from Priyanka Dwivedi\nfirst repo: https://github.com/priya-dwivedi/aerial_pedestrian_detection/blob/master/README.md\nsecond sharing and in Turkey there is a contest Teknofest and Sıddık Acıl team used RetinaNet image classification but this repo is Turkish, maybe benefit Turkish users\nsecond repo: https://github.com/huma-teknofest/Keras-RetinaNet-for-Teknofest-2019", "votes": null } ], "comments": [ { "id": 414489, "author_name": "sgalib", "author_url": "", "post_date": "11/02/2018 21:25:44", "content": "Updated. Actually one of Branden's model lb was 0.203 in public lb. I did not check mine individually during stage 1. However, still some of my individual model scored around 0.20 in stage 2 private lb.
", "votes": null, "replies": [ { "id": 414492, "author_name": "mpsampat", "author_url": "", "post_date": "11/02/2018 21:47:10", "content": "Awesome! Thanks Shai! \nOnce more people add the parameters I will make a git repo like this one:
\n\n\n\nI made this one for Mask-RCNN after DSB_2018.
", "votes": null, "replies": [] }, { "id": 414498, "author_name": "hxy777", "author_url": "", "post_date": "11/02/2018 22:20:51", "content": "Thanks a lot for sharing your parameters! I just used a single RetinaNet for this competition and scored around 0.19 in private lb.
", "votes": null, "replies": [] }, { "id": 416063, "author_name": "mdlszhengli", "author_url": "", "post_date": "11/06/2018 05:34:30", "content": "If i don't see this result, I certainly won't believe that ensemble has such a good effect. Thanks for sharing.
", "votes": null, "replies": [] } ] }, { "id": 414521, "author_name": "taindow", "author_url": "", "post_date": "11/02/2018 23:10:20", "content": "Thanks for sharing guys, added mine too.
", "votes": null, "replies": [] }, { "id": 414535, "author_name": "mpsampat", "author_url": "", "post_date": "11/02/2018 23:32:59", "content": "@Shai, @hxy777 and @taindow : thanks so much for filling out the google sheet :)
", "votes": null, "replies": [] }, { "id": 416798, "author_name": "felipekitamura", "author_url": "", "post_date": "11/07/2018 09:54:44", "content": "Nice initiative, @Mehul. Our team is going to publish our solution later this week. I have started filling some parameters in your google doc, though.
", "votes": null, "replies": [] }, { "id": 745133, "author_name": "alistnel", "author_url": "", "post_date": "02/13/2020 14:37:37", "content": "it's been a long time but i have 2 good resources.\nFirst, Stanford Drone Dataset from Priyanka Dwivedi\nfirst repo: https://github.com/priya-dwivedi/aerial_pedestrian_detection/blob/master/README.md\nsecond sharing and in Turkey there is a contest Teknofest and Sıddık Acıl team used RetinaNet image classification but this repo is Turkish, maybe benefit Turkish users\nsecond repo: https://github.com/huma-teknofest/Keras-RetinaNet-for-Teknofest-2019
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "414023": "Hi All! \nThere were couple of very useful discussions on using RetinaNet implemented in Keras\nhttps://github.com/fizyr/keras-retinanet/\nI started late with it and could only run a few experiments with it. Some of my parameters and results for stage-1 are here:\nhttps://docs.google.com/spreadsheets/d/18ln0yxA19OZ4q6Pf780LD4r8BYhRKZ2xgZ3aQpcJ2uw/edit?usp=sharing\n\nI recall some users had great results. @Shai you got 0.203 with a single retina-net model right ? \nIf you used RetinaNet, could you please add your parameters to this google doc ? It would help everyone learn! \nThanks for all the great discussions you all! :) \nMehul", "414489": "Updated. Actually one of Branden's model lb was 0.203 in public lb. I did not check mine individually during stage 1. However, still some of my individual model scored around 0.20 in stage 2 private lb.", "414492": "Awesome! Thanks Shai! \nOnce more people add the parameters I will make a git repo like this one: \n\nhttps://github.com/mpsampat/DSB_2018_Parameter_Selection_For_Matterport_Mask_RCNN/blob/master/comparison_of_Matterport_Mask_RCNN_parameters.md\n\nI made this one for Mask-RCNN after DSB_2018.", "414498": "Thanks a lot for sharing your parameters! I just used a single RetinaNet for this competition and scored around 0.19 in private lb.", "414521": "Thanks for sharing guys, added mine too.", "414535": "Shai, @hxy777 and @taindow : thanks so much for filling out the google sheet :)", "416063": "If i don't see this result, I certainly won't believe that ensemble has such a good effect. Thanks for sharing.", "416798": "Nice initiative, @Mehul. Our team is going to publish our solution later this week. I have started filling some parameters in your google doc, though.", "745133": "it's been a long time but i have 2 good resources.\nFirst, Stanford Drone Dataset from Priyanka Dwivedi\nfirst repo: https://github.com/priya-dwivedi/aerial_pedestrian_detection/blob/master/README.md\nsecond sharing and in Turkey there is a contest Teknofest and Sıddık Acıl team used RetinaNet image classification but this repo is Turkish, maybe benefit Turkish users\nsecond repo: https://github.com/huma-teknofest/Keras-RetinaNet-for-Teknofest-2019" }, "source": "meta" }