![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4310004%2F9012ac8646c313df1f2c619de4afbebd%2Finbox_10714081_bfea50c01ca6d057bb4a4fb5a49f8eae_Screenshot%202025-03-21%20075234.png?generation=1743299940297729&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4310004%2F0bc20ecf1194fe2d49725264d2c0db31%2F__results___24_1.png?generation=1743397563369189&alt=media\")
We always see you solving projects with GNN (or other models based on graph construction)—truly inspiring! May I ask what advantages GNN offers in this competition? Thanks！
", "rawMarkdown": "We always see you solving projects with GNN (or other models based on graph construction)—truly inspiring! May I ask what advantages GNN offers in this competition? Thanks！", "votes": null }, { "id": "3163155", "postDate": "03/30/2025 11:15:16", "content": "@gygyygygy I plan to model the structure of bacteria cause motor usually exists in the head of it. So if I have a structural graph, I can easily find the head and determine whether there's a motor existed. Here's another result.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4310004%2F7c1488dd0760efe5e8ac342ed399f902%2Finbox_10714081_bfea50c01ca6d057bb4a4fb5a49f8eae_Screenshot%202025-03-21%20075234.png?generation=1743333425548673&alt=media\")
Cool !!
\nAn edge detection algorithm followed by a dilation does the job well. Since edges are continuous apply DFS to traverse the boundary of the motor.
I'd be really interested in seeing your code once this competition is done! It seems like a high potential vein of research going forward.
", "rawMarkdown": "I'd be really interested in seeing your code once this competition is done! It seems like a high potential vein of research going forward.", "votes": null }, { "id": "3167873", "postDate": "04/01/2025 22:32:14", "content": "@andrewjdarley The preprocessing is quit fancy since there are a lot of noises in the images. But actually this idea is a very good research topic cause this method is completely different to object detection.
", "rawMarkdown": "andrewjdarley The preprocessing is quit fancy since there are a lot of noises in the images. But actually this idea is a very good research topic cause this method is completely different to object detection.", "votes": null }, { "id": "3167876", "postDate": "04/01/2025 22:34:28", "content": "@vigneshwar472 yeah the location of the motor always locate in the edge, this gives another direction to handle this comp.
", "rawMarkdown": "vigneshwar472 yeah the location of the motor always locate in the edge, this gives another direction to handle this comp.", "votes": null }, { "id": "3169550", "postDate": "04/03/2025 15:24:44", "content": "Is your current LB purely based off of graph?
", "rawMarkdown": "Is your current LB purely based off of graph?", "votes": null }, { "id": "3169861", "postDate": "04/04/2025 01:37:29", "content": "@cody11null yeah. But I haven't optimized it.
", "rawMarkdown": "cody11null yeah. But I haven't optimized it.", "votes": null }, { "id": "3169867", "postDate": "04/04/2025 01:50:12", "content": "Very impressive! Great work!
", "rawMarkdown": "Very impressive! Great work!", "votes": null }, { "id": "3193169", "postDate": "05/04/2025 01:58:54", "content": "Your approach is an exciting and different approach from the norm. Thumbs up for you.
", "rawMarkdown": "Your approach is an exciting and different approach from the norm. Thumbs up for you.", "votes": null }, { "id": "3193171", "postDate": "05/04/2025 02:16:17", "content": "So here I show some different graphs in 3d expansions which are my current best solutions. The differences between them are the densities. I observe this is critical for this type of data.
\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4310004%2Fe6957e319f1087629d3eef5c64c86200%2Fknn.png?generation=1746324842424889&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4310004%2Fd8d7e5bd71efdebd15299e43148837d0%2Fradius.png?generation=1746324855935847&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4310004%2F5e09c7e8ac63cc2ae2efc1beff88c245%2Fdelaunay.png?generation=1746324874709518&alt=media\")
At this point I am wishing I forced myself to try graph haha
", "rawMarkdown": "At this point I am wishing I forced myself to try graph haha", "votes": null }, { "id": "3193387", "postDate": "05/04/2025 09:50:27", "content": "May I know if the result of training with GNN is stable? Since my experience with YOLO is not, so if it is with GNN, I will definitely try it.
", "rawMarkdown": "May I know if the result of training with GNN is stable? Since my experience with YOLO is not, so if it is with GNN, I will definitely try it.", "votes": null } ], "comments": [ { "id": 3162999, "author_name": "gygyygygy", "author_url": "", "post_date": "03/30/2025 06:49:25", "content": "We always see you solving projects with GNN (or other models based on graph construction)—truly inspiring! May I ask what advantages GNN offers in this competition? Thanks！
", "votes": null, "replies": [ { "id": 3163155, "author_name": "tom99763", "author_url": "", "post_date": "03/30/2025 11:15:16", "content": "@gygyygygy I plan to model the structure of bacteria cause motor usually exists in the head of it. So if I have a structural graph, I can easily find the head and determine whether there's a motor existed. Here's another result.
\n
Cool !!
\nAn edge detection algorithm followed by a dilation does the job well. Since edges are continuous apply DFS to traverse the boundary of the motor.
@vigneshwar472 yeah the location of the motor always locate in the edge, this gives another direction to handle this comp.
", "votes": null, "replies": [] } ] } ] } ] }, { "id": 3167805, "author_name": "andrewjdarley", "author_url": "", "post_date": "04/01/2025 20:56:50", "content": "I'd be really interested in seeing your code once this competition is done! It seems like a high potential vein of research going forward.
", "votes": null, "replies": [ { "id": 3167873, "author_name": "tom99763", "author_url": "", "post_date": "04/01/2025 22:32:14", "content": "@andrewjdarley The preprocessing is quit fancy since there are a lot of noises in the images. But actually this idea is a very good research topic cause this method is completely different to object detection.
", "votes": null, "replies": [] } ] }, { "id": 3169550, "author_name": "cody11null", "author_url": "", "post_date": "04/03/2025 15:24:44", "content": "Is your current LB purely based off of graph?
", "votes": null, "replies": [ { "id": 3169861, "author_name": "tom99763", "author_url": "", "post_date": "04/04/2025 01:37:29", "content": "@cody11null yeah. But I haven't optimized it.
", "votes": null, "replies": [ { "id": 3169867, "author_name": "cody11null", "author_url": "", "post_date": "04/04/2025 01:50:12", "content": "Very impressive! Great work!
", "votes": null, "replies": [] } ] } ] }, { "id": 3193169, "author_name": "ynhuhu", "author_url": "", "post_date": "05/04/2025 01:58:54", "content": "Your approach is an exciting and different approach from the norm. Thumbs up for you.
", "votes": null, "replies": [] }, { "id": 3193171, "author_name": "tom99763", "author_url": "", "post_date": "05/04/2025 02:16:17", "content": "So here I show some different graphs in 3d expansions which are my current best solutions. The differences between them are the densities. I observe this is critical for this type of data.
\n
At this point I am wishing I forced myself to try graph haha
", "votes": null, "replies": [] } ] }, { "id": 3193387, "author_name": "levinguyen02", "author_url": "", "post_date": "05/04/2025 09:50:27", "content": "May I know if the result of training with GNN is stable? Since my experience with YOLO is not, so if it is with GNN, I will definitely try it.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3162910": "\n\n\n\n### [Example notebook](https://www.kaggle.com/code/tom99763/gnn-example-byu)\n\n### [Submission notebook](https://www.kaggle.com/code/tom99763/tomv3-gnn-submission-byu)\n\n### [Preprocess notebook](https://www.kaggle.com/code/tom99763/first-idea-to-extract-keypoints-byu)", "3162999": "We always see you solving projects with GNN (or other models based on graph construction)—truly inspiring! May I ask what advantages GNN offers in this competition? Thanks！", "3163155": "gygyygygy I plan to model the structure of bacteria cause motor usually exists in the head of it. So if I have a structural graph, I can easily find the head and determine whether there's a motor existed. Here's another result.\n\n", "3167712": "Cool !! \nAn edge detection algorithm followed by a dilation does the job well. Since edges are continuous apply DFS to traverse the boundary of the motor.", "3167805": "I'd be really interested in seeing your code once this competition is done! It seems like a high potential vein of research going forward.", "3167873": "andrewjdarley The preprocessing is quit fancy since there are a lot of noises in the images. But actually this idea is a very good research topic cause this method is completely different to object detection.", "3167876": "vigneshwar472 yeah the location of the motor always locate in the edge, this gives another direction to handle this comp.", "3169550": "Is your current LB purely based off of graph?", "3169861": "cody11null yeah. But I haven't optimized it.", "3169867": "Very impressive! Great work!", "3193169": "Your approach is an exciting and different approach from the norm. Thumbs up for you.", "3193171": "So here I show some different graphs in 3d expansions which are my current best solutions. The differences between them are the densities. I observe this is critical for this type of data.\n\n\n\n\n\n", "3193239": "At this point I am wishing I forced myself to try graph haha", "3193387": "May I know if the result of training with GNN is stable? Since my experience with YOLO is not, so if it is with GNN, I will definitely try it." }, "source": "meta" }