{ "id": 310779, "title": "cropped dataset using TokenCut", "url": "/competitions/happy-whale-and-dolphin/discussion/310779", "author_name": "k_s", "post_date": "2022-03-03T05:34:21.138000", "votes": 26, "comment_count": 2, "views": 0, "content": "

Dataset Link: cropped dataset using tokencut
\nTFRecords Version : cropped tfrecords dataset using tokencut
\nTokenCut Project Page
\n\"cropped

\n

From some experiments I have done, I think it is important to use cropped data.
\nThank you, @phalanx and @awsaf49, for publishing the useful dataset.
\n@phalanx's cropped dataset using detic and @awsaf49's cropped dataset using YOLOv5
\nI thought it would be a good opportunity to try using TokenCut, which was recently released, so I made a cropped dataset. I used @rdizzl3's 384x384 image as input and used vit_base as a model architecture.

\n

Cropped TFRecords Dataset
\nI also converted all of the data, including @phalanx 's and @awsaf49's, into tfrecords.
\nThe link to the dataset is here.
\ncropped tfrecords dataset using detic
\ncropped tfrecords dataset using yolov5 conf_threshold=0.05
\ncropped tfrecords dataset using yolov5 conf_threshold=0.10

\n

Data Visualization
\nFinally, I also published a notebook to compare these datasets.
\nhappywhale_data_visualization Notebook

\n

I hope you find it useful.

", "messages": [ { "id": 1710489, "postDate": "2022-03-03T05:34:21.140Z", "content": "

Dataset Link: cropped dataset using tokencut
\nTFRecords Version : cropped tfrecords dataset using tokencut
\nTokenCut Project Page
\n\"cropped

\n

From some experiments I have done, I think it is important to use cropped data.
\nThank you, @phalanx and @awsaf49, for publishing the useful dataset.
\n@phalanx's cropped dataset using detic and @awsaf49's cropped dataset using YOLOv5
\nI thought it would be a good opportunity to try using TokenCut, which was recently released, so I made a cropped dataset. I used @rdizzl3's 384x384 image as input and used vit_base as a model architecture.

\n

Cropped TFRecords Dataset
\nI also converted all of the data, including @phalanx 's and @awsaf49's, into tfrecords.
\nThe link to the dataset is here.
\ncropped tfrecords dataset using detic
\ncropped tfrecords dataset using yolov5 conf_threshold=0.05
\ncropped tfrecords dataset using yolov5 conf_threshold=0.10

\n

Data Visualization
\nFinally, I also published a notebook to compare these datasets.
\nhappywhale_data_visualization Notebook

\n

I hope you find it useful.

", "rawMarkdown": "Dataset Link: [cropped dataset using tokencut](https://www.kaggle.com/ksork6s4/happywhalecroppeddatasettokencut)\nTFRecords Version : [cropped tfrecords dataset using tokencut](https://www.kaggle.com/ksork6s4/happywhale-tfrecords-tokencut-384)\n[TokenCut Project Page](https://www.m-psi.fr/Papers/TokenCut2022/)\n![cropped images](https://i.postimg.cc/6qbbzsxn/2022-03-03-141300.png)\n\nFrom some experiments I have done, I think it is important to use cropped data.\nThank you, @phalanx and @awsaf49, for publishing the useful dataset.\n@phalanx's [cropped dataset using detic](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/305503) and @awsaf49's [cropped dataset using YOLOv5](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/305942)\nI thought it would be a good opportunity to try using TokenCut, which was recently released, so I made a cropped dataset. I used @rdizzl3's [384x384 image](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/304686) as input and used vit_base as a model architecture.\n\n\n\n**Cropped TFRecords Dataset**\nI also converted all of the data, including @phalanx 's and @awsaf49's, into tfrecords.\nThe link to the dataset is here.\n[cropped tfrecords dataset using detic](https://www.kaggle.com/ksork6s4/happywhale-tfrecords-detic-512)\n[cropped tfrecords dataset using yolov5 conf_threshold=0.05](https://www.kaggle.com/ksork6s4/happywhale-tfrecords-yolov5-512-005)\n[cropped tfrecords dataset using yolov5 conf_threshold=0.10](https://www.kaggle.com/ksork6s4/happywhale-tfrecords-yolov5-512-010)\n\n\n\n**Data Visualization**\nFinally, I also published a notebook to compare these datasets.\n[happywhale_data_visualization Notebook](https://www.kaggle.com/ksork6s4/happywhale-data-visualization/notebook)\n\n\nI hope you find it useful.", "votes": 26 }, { "id": 1711502, "postDate": "2022-03-04T04:13:23.770Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 1710675, "postDate": "2022-03-03T08:21:56.007Z", "content": "

thanks for sharing. upvote.

", "rawMarkdown": "thanks for sharing. upvote." } ], "comments": [ { "id": 1711502, "author_name": "", "author_url": "", "post_date": "2022-03-04T04:13:23.770000", "content": "", "votes": 1, "replies": [] }, { "id": 1710675, "author_name": "dragon zhang", "author_url": "", "post_date": "2022-03-03T08:21:56.007000", "content": "

thanks for sharing. upvote.

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1710489": "Dataset Link: [cropped dataset using tokencut](https://www.kaggle.com/ksork6s4/happywhalecroppeddatasettokencut)\nTFRecords Version : [cropped tfrecords dataset using tokencut](https://www.kaggle.com/ksork6s4/happywhale-tfrecords-tokencut-384)\n[TokenCut Project Page](https://www.m-psi.fr/Papers/TokenCut2022/)\n![cropped images](https://i.postimg.cc/6qbbzsxn/2022-03-03-141300.png)\n\nFrom some experiments I have done, I think it is important to use cropped data.\nThank you, @phalanx and @awsaf49, for publishing the useful dataset.\n@phalanx's [cropped dataset using detic](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/305503) and @awsaf49's [cropped dataset using YOLOv5](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/305942)\nI thought it would be a good opportunity to try using TokenCut, which was recently released, so I made a cropped dataset. I used @rdizzl3's [384x384 image](https://www.kaggle.com/c/happy-whale-and-dolphin/discussion/304686) as input and used vit_base as a model architecture.\n\n\n\n**Cropped TFRecords Dataset**\nI also converted all of the data, including @phalanx 's and @awsaf49's, into tfrecords.\nThe link to the dataset is here.\n[cropped tfrecords dataset using detic](https://www.kaggle.com/ksork6s4/happywhale-tfrecords-detic-512)\n[cropped tfrecords dataset using yolov5 conf_threshold=0.05](https://www.kaggle.com/ksork6s4/happywhale-tfrecords-yolov5-512-005)\n[cropped tfrecords dataset using yolov5 conf_threshold=0.10](https://www.kaggle.com/ksork6s4/happywhale-tfrecords-yolov5-512-010)\n\n\n\n**Data Visualization**\nFinally, I also published a notebook to compare these datasets.\n[happywhale_data_visualization Notebook](https://www.kaggle.com/ksork6s4/happywhale-data-visualization/notebook)\n\n\nI hope you find it useful.", "1711502": "", "1710675": "thanks for sharing. upvote." } }