{ "id": 309714, "title": "The Sword of a Thousand Truths - Finding Free Willy w/ YOLOv5", "url": "/competitions/happy-whale-and-dolphin/discussion/309714", "author_name": "", "post_date": "2022-02-25T02:04:36.066404300Z", "votes": 13, "comment_count": 9, "views": 0, "content": "

Topic Title exaggeration aside….😃. I tagged ~2200 images and got ~1500 tags from @bsridatta link: https://www.kaggle.com/bsridatta/happierwhale-1500-manual-annotations
\nand ran through yolov5.

\n

BEST PT FILE: https://www.kaggle.com/click2clone/happywhale-sword-of-thousand-truths

\n

I haven't tried to share anything yet, and I'm still learning this site, so if it doesn't work let me know and I'll try and fix it. It's pretty big but inference time isn't too bad.

\n

One of the images that my other yolo training attempts got wrong is the below, cause of all the crazy tree detail, it kept finding whales in trees, so I was happy to see it got it right:

\n

", "messages": [ { "id": "1703910", "postDate": "02/25/2022 02:04:36", "content": "

Topic Title exaggeration aside….😃. I tagged ~2200 images and got ~1500 tags from @bsridatta link: https://www.kaggle.com/bsridatta/happierwhale-1500-manual-annotations
\nand ran through yolov5.

\n

BEST PT FILE: https://www.kaggle.com/click2clone/happywhale-sword-of-thousand-truths

\n

I haven't tried to share anything yet, and I'm still learning this site, so if it doesn't work let me know and I'll try and fix it. It's pretty big but inference time isn't too bad.

\n

One of the images that my other yolo training attempts got wrong is the below, cause of all the crazy tree detail, it kept finding whales in trees, so I was happy to see it got it right:

\n

", "rawMarkdown": "Topic Title exaggeration aside....😃. I tagged ~2200 images and got ~1500 tags from @bsridatta link: https://www.kaggle.com/bsridatta/happierwhale-1500-manual-annotations\nand ran through yolov5.\n\nBEST PT FILE: https://www.kaggle.com/click2clone/happywhale-sword-of-thousand-truths\n\nI haven't tried to share anything yet, and I'm still learning this site, so if it doesn't work let me know and I'll try and fix it. It's pretty big but inference time isn't too bad.\n\nOne of the images that my other yolo training attempts got wrong is the below, cause of all the crazy tree detail, it kept finding whales in trees, so I was happy to see it got it right:\n\n", "votes": null }, { "id": "1704009", "postDate": "02/25/2022 04:39:20", "content": "

Wow, awesome work, @click2clone ! 3700 is a good amount of training data. Hopefully the images didn't overlap. The detection is also pretty neat. The model size shouldn't matter , ideally it's a one time thing to do. And this step is the most important one.

", "rawMarkdown": "Wow, awesome work, @click2clone ! 3700 is a good amount of training data. Hopefully the images didn't overlap. The detection is also pretty neat. The model size shouldn't matter , ideally it's a one time thing to do. And this step is the most important one.", "votes": null }, { "id": "1704296", "postDate": "02/25/2022 11:41:18", "content": "

Great work @click2clone! Are you planning on releasing the annotations you made to a dataset?

", "rawMarkdown": "Great work @click2clone! Are you planning on releasing the annotations you made to a dataset?", "votes": null }, { "id": "1704537", "postDate": "02/25/2022 15:56:28", "content": "

Great work

", "rawMarkdown": "Great work", "votes": null }, { "id": "1704671", "postDate": "02/25/2022 17:46:38", "content": "

Yes, I meant to, but it took so long to upload that pytorch model, i guess i forgot, so i added it to the link above.

", "rawMarkdown": "Yes, I meant to, but it took so long to upload that pytorch model, i guess i forgot, so i added it to the link above.", "votes": null }, { "id": "1704675", "postDate": "02/25/2022 17:48:03", "content": "

Thank you, half is from @bsridatta though

", "rawMarkdown": "Thank you, half is from @bsridatta though", "votes": null }, { "id": "1704679", "postDate": "02/25/2022 17:51:20", "content": "

I posted all the txt files but almost half are yours, so I'm feeling uneasy. If you want to copy them all and repost, I'll delete them on my end. I don't want to take any credit for it all.

", "rawMarkdown": "I posted all the txt files but almost half are yours, so I'm feeling uneasy. If you want to copy them all and repost, I'll delete them on my end. I don't want to take any credit for it all.", "votes": null }, { "id": "1704685", "postDate": "02/25/2022 17:54:03", "content": "

Thank you!

", "rawMarkdown": "Thank you!", "votes": null }, { "id": "1704707", "postDate": "02/25/2022 18:14:15", "content": "

No worries @click2clone :) nice that you are thinking about it. it's perfectly fine as you are mentioning me or the notebook wherever you used them. I made it public so others can use. you seem to be also training a model on more data, I might not. If you can share that with everyone it would be great. Thanks!

", "rawMarkdown": "No worries @click2clone :) nice that you are thinking about it. it's perfectly fine as you are mentioning me or the notebook wherever you used them. I made it public so others can use. you seem to be also training a model on more data, I might not. If you can share that with everyone it would be great. Thanks!", "votes": null }, { "id": "1704764", "postDate": "02/25/2022 19:16:32", "content": "

Nice! it seems great!

", "rawMarkdown": "Nice! it seems great!", "votes": null } ], "comments": [ { "id": 1704009, "author_name": "bsridatta", "author_url": "", "post_date": "02/25/2022 04:39:20", "content": "

Wow, awesome work, @click2clone ! 3700 is a good amount of training data. Hopefully the images didn't overlap. The detection is also pretty neat. The model size shouldn't matter , ideally it's a one time thing to do. And this step is the most important one.

", "votes": null, "replies": [ { "id": 1704679, "author_name": "click2clone", "author_url": "", "post_date": "02/25/2022 17:51:20", "content": "

I posted all the txt files but almost half are yours, so I'm feeling uneasy. If you want to copy them all and repost, I'll delete them on my end. I don't want to take any credit for it all.

", "votes": null, "replies": [] }, { "id": 1704707, "author_name": "bsridatta", "author_url": "", "post_date": "02/25/2022 18:14:15", "content": "

No worries @click2clone :) nice that you are thinking about it. it's perfectly fine as you are mentioning me or the notebook wherever you used them. I made it public so others can use. you seem to be also training a model on more data, I might not. If you can share that with everyone it would be great. Thanks!

", "votes": null, "replies": [] } ] }, { "id": 1704296, "author_name": "vexxingbanana", "author_url": "", "post_date": "02/25/2022 11:41:18", "content": "

Great work @click2clone! Are you planning on releasing the annotations you made to a dataset?

", "votes": null, "replies": [ { "id": 1704671, "author_name": "click2clone", "author_url": "", "post_date": "02/25/2022 17:46:38", "content": "

Yes, I meant to, but it took so long to upload that pytorch model, i guess i forgot, so i added it to the link above.

", "votes": null, "replies": [] }, { "id": 1704685, "author_name": "vexxingbanana", "author_url": "", "post_date": "02/25/2022 17:54:03", "content": "

Thank you!

", "votes": null, "replies": [] } ] }, { "id": 1704537, "author_name": "sulaimanmutawalli", "author_url": "", "post_date": "02/25/2022 15:56:28", "content": "

Great work

", "votes": null, "replies": [ { "id": 1704675, "author_name": "click2clone", "author_url": "", "post_date": "02/25/2022 17:48:03", "content": "

Thank you, half is from @bsridatta though

", "votes": null, "replies": [] } ] }, { "id": 1704764, "author_name": "deepkim", "author_url": "", "post_date": "02/25/2022 19:16:32", "content": "

Nice! it seems great!

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1703910": "Topic Title exaggeration aside....😃. I tagged ~2200 images and got ~1500 tags from @bsridatta link: https://www.kaggle.com/bsridatta/happierwhale-1500-manual-annotations\nand ran through yolov5.\n\nBEST PT FILE: https://www.kaggle.com/click2clone/happywhale-sword-of-thousand-truths\n\nI haven't tried to share anything yet, and I'm still learning this site, so if it doesn't work let me know and I'll try and fix it. It's pretty big but inference time isn't too bad.\n\nOne of the images that my other yolo training attempts got wrong is the below, cause of all the crazy tree detail, it kept finding whales in trees, so I was happy to see it got it right:\n\n", "1704009": "Wow, awesome work, @click2clone ! 3700 is a good amount of training data. Hopefully the images didn't overlap. The detection is also pretty neat. The model size shouldn't matter , ideally it's a one time thing to do. And this step is the most important one.", "1704296": "Great work @click2clone! Are you planning on releasing the annotations you made to a dataset?", "1704537": "Great work", "1704671": "Yes, I meant to, but it took so long to upload that pytorch model, i guess i forgot, so i added it to the link above.", "1704675": "Thank you, half is from @bsridatta though", "1704679": "I posted all the txt files but almost half are yours, so I'm feeling uneasy. If you want to copy them all and repost, I'll delete them on my end. I don't want to take any credit for it all.", "1704685": "Thank you!", "1704707": "No worries @click2clone :) nice that you are thinking about it. it's perfectly fine as you are mentioning me or the notebook wherever you used them. I made it public so others can use. you seem to be also training a model on more data, I might not. If you can share that with everyone it would be great. Thanks!", "1704764": "Nice! it seems great!" }, "source": "meta" }