The more you invest in this competition, the more you will realize that the dataset and ways you are handling it will play a crucial role.
\nA lot of great work has already been done and this post is a way to appreciate their works and amass them here.
\nThe more you invest in this competition, the more you will realize that the dataset and ways you are handling it will play a crucial role.
\nA lot of great work has already been done and this post is a way to appreciate their works and amass them here.
\n@ayuraj thank you for mentioning my work. I am still working on better solution - removing background (now it is almost acceptable - almost). To improve it even more I need retrain model - it requires annotations like in semantic segmentation. It is time consuming but … now testing simillarity model performance (using my dataset). I am almost sure that there is no huge benefit from removing background - still testing. If you prepare good data there is no need to remove background. I will let you know when finish testing.
", "rawMarkdown": "ayuraj thank you for mentioning my work. I am still working on better solution - removing background (now it is almost acceptable - almost). To improve it even more I need retrain model - it requires annotations like in semantic segmentation. It is time consuming but ... now testing simillarity model performance (using my dataset). I am almost sure that there is no huge benefit from removing background - still testing. If you prepare good data there is no need to remove background. I will let you know when finish testing.", "votes": null }, { "id": "1703825", "postDate": "02/24/2022 23:19:30", "content": "@remekkinas I found this segmentation dataset, you may be able to use it.
\nI haven't checked it yet. I think mostly dolphins, but I am not sure.
https://data.ncl.ac.uk/articles/dataset/NDD20_zip/12357383
", "rawMarkdown": "remekkinas I found this segmentation dataset, you may be able to use it.\nI haven't checked it yet. I think mostly dolphins, but I am not sure.\n\nhttps://data.ncl.ac.uk/articles/dataset/NDD20_zip/12357383", "votes": null }, { "id": "1704305", "postDate": "02/25/2022 11:50:37", "content": "Thank you for putting this list together! I have also manually labelled 1,680 images with camera viewpoints in case it helps!
\n", "rawMarkdown": "Thank you for putting this list together! I have also manually labelled 1,680 images with camera viewpoints in case it helps!\n- [Happywhale Camera Viewpoints Dataset](https://www.kaggle.com/frlemarchand/happywhale-camera-viewpoints)", "votes": null }, { "id": "1704812", "postDate": "02/25/2022 20:53:49", "content": "Thanks for sharing @frlemarchand. Updated the list.
", "rawMarkdown": "Thanks for sharing @frlemarchand. Updated the list.", "votes": null }, { "id": "1704844", "postDate": "02/25/2022 22:01:31", "content": "Thank you @pestipeti!
\nSegmentation should work in my case. I have to improve NN a little bit and had to take decision -> do it manually or find dataset. Now problem looks easier for me :) Thank you!
@ayuraj thank you for mentioning my work. I am still working on better solution - removing background (now it is almost acceptable - almost). To improve it even more I need retrain model - it requires annotations like in semantic segmentation. It is time consuming but … now testing simillarity model performance (using my dataset). I am almost sure that there is no huge benefit from removing background - still testing. If you prepare good data there is no need to remove background. I will let you know when finish testing.
", "votes": null, "replies": [ { "id": 1703825, "author_name": "pestipeti", "author_url": "", "post_date": "02/24/2022 23:19:30", "content": "@remekkinas I found this segmentation dataset, you may be able to use it.
\nI haven't checked it yet. I think mostly dolphins, but I am not sure.
https://data.ncl.ac.uk/articles/dataset/NDD20_zip/12357383
", "votes": null, "replies": [] }, { "id": 1704844, "author_name": "remekkinas", "author_url": "", "post_date": "02/25/2022 22:01:31", "content": "Thank you @pestipeti!
\nSegmentation should work in my case. I have to improve NN a little bit and had to take decision -> do it manually or find dataset. Now problem looks easier for me :) Thank you!
Thank you for putting this list together! I have also manually labelled 1,680 images with camera viewpoints in case it helps!
\n", "votes": null, "replies": [ { "id": 1704812, "author_name": "ayuraj", "author_url": "", "post_date": "02/25/2022 20:53:49", "content": "Thanks for sharing @frlemarchand. Updated the list.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1703798": "The more you invest in this competition, the more you will realize that the dataset and ways you are handling it will play a crucial role. \n\nA lot of great work has already been done and this post is a way to appreciate their works and amass them here. \n\n## Raw Images Resized\n\n* [Reduced Resolution Image Data (128 x 128, 256 x 256, 384 x 384) 🐋](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/304686) by @rdizzl3 - [128x128](https://www.kaggle.com/rdizzl3/jpeg-happywhale-128x128), [256x256](https://www.kaggle.com/rdizzl3/jpeg-happywhale-256x256), [384x384](https://www.kaggle.com/rdizzl3/jpeg-happywhale-384x384)\n\n## TFRecords of Raw Images\n\n* [HappyWhale TFRecords](https://www.kaggle.com/ks2019/happywhale-tfrecords) by @ks2019 - [Dataset](https://www.kaggle.com/ks2019/happywhale-tfrecords-v1)\n\n## Bounding Box Based Crops\n\n* [Happywhale: Cropped Dataset [YOLOv5] ✂️](https://www.kaggle.com/awsaf49/happywhale-cropped-dataset-yolov5/) by @awsaf49 - [Dataset](https://www.kaggle.com/awsaf49/happywhale-cropped-dataset-yolov5-ds)\n* [cropped&resized(512x512) dataset using detic](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/305503) by @phalanx - [Dataset](https://www.kaggle.com/phalanx/whale2-cropped-dataset)\n* [TF records with Detic Crop](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/309611) by @ragnar123 \n\n## Augmented Images\n\n* [🐋🐬Happywhale: Augmented](https://www.kaggle.com/ruchi798/happywhale-augmented) by @ruchi798 \n\n## Segmentation Masks\n\n* [HappyWhales LabelMe Segmentation Dataset](https://www.kaggle.com/shubhambaid/happywhales-labelme-segmentation-dataset) by @shubhambaid (not full dataset)\n\n## Teased us :P\n\n* [🔥 DATASET - dorsal fins for all IDs without background 🔥](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/309214) by @remekkinas \n\n## Dataset Painpoints and EDA\n\n* [🐧 Things to know before starting image preprocessing](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/308026) by @andradaolteanu.\n\n## Camera View Point\n\n* [Happywhale Camera Viewpoints Dataset](https://www.kaggle.com/frlemarchand/happywhale-camera-viewpoints)", "1703813": "ayuraj thank you for mentioning my work. I am still working on better solution - removing background (now it is almost acceptable - almost). To improve it even more I need retrain model - it requires annotations like in semantic segmentation. It is time consuming but ... now testing simillarity model performance (using my dataset). I am almost sure that there is no huge benefit from removing background - still testing. If you prepare good data there is no need to remove background. I will let you know when finish testing.", "1703825": "remekkinas I found this segmentation dataset, you may be able to use it.\nI haven't checked it yet. I think mostly dolphins, but I am not sure.\n\nhttps://data.ncl.ac.uk/articles/dataset/NDD20_zip/12357383", "1704305": "Thank you for putting this list together! I have also manually labelled 1,680 images with camera viewpoints in case it helps!\n- [Happywhale Camera Viewpoints Dataset](https://www.kaggle.com/frlemarchand/happywhale-camera-viewpoints)", "1704812": "Thanks for sharing @frlemarchand. Updated the list.", "1704844": "Thank you @pestipeti! \nSegmentation should work in my case. I have to improve NN a little bit and had to take decision -> do it manually or find dataset. Now problem looks easier for me :) Thank you!" }, "source": "meta" }