![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F7492649%2F5033f0a0c5e4b735700a661a6e76929e%2F_20240126153731.png?generation=1706254675067990&alt=media\")
This might be a straightforward question, but I’m curious to know based on your experiments, is the public test closer to densely segmented or sparsely segmented? Thank you！
", "messages": [ { "id": "2620515", "postDate": "01/26/2024 07:20:18", "content": "This might be a straightforward question, but I’m curious to know based on your experiments, is the public test closer to densely segmented or sparsely segmented? Thank you！
", "rawMarkdown": "This might be a straightforward question, but I’m curious to know based on your experiments, is the public test closer to densely segmented or sparsely segmented? Thank you！", "votes": null }, { "id": "2620540", "postDate": "01/26/2024 07:38:12", "content": "Currently, my personal experiment process for exploring density is as follows: 1) Find the best validation dataset 2) Find the best training data. My conclusion is that dense1 and dense3 are the closest to the public test dataset. However, I’m not sure if it’s a problem with the experimental setup, so I’d like to discuss this with everyone.
Hi. In my opinion. If you are scoring the performance on segmentation. They should be as dense as they could.
", "rawMarkdown": "Hi. In my opinion. If you are scoring the performance on segmentation. They should be as dense as they could.", "votes": null }, { "id": "2620597", "postDate": "01/26/2024 08:30:58", "content": "I have another question: Are the test datasets sliced along the z-axis？
", "rawMarkdown": "I have another question: Are the test datasets sliced along the z-axis？", "votes": null }, { "id": "2620600", "postDate": "01/26/2024 08:33:23", "content": "Thank you！It makes sense! I have carefully read the process of maual annotation https://arxiv.org/pdf/2311.13319.pdf
", "rawMarkdown": "Thank you！It makes sense! I have carefully read the process of maual annotation [https://arxiv.org/pdf/2311.13319.pdf](url)", "votes": null }, { "id": "2620734", "postDate": "01/26/2024 10:26:57", "content": "https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/460833
", "rawMarkdown": "https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/460833", "votes": null }, { "id": "2623124", "postDate": "01/28/2024 01:59:48", "content": "i train my model using d1dense and val on d3dense, local surfacedice is 0.857,but lb is only 0.84, and the public 0.859, i test on d3 dense with local surface dice only 0.806.😥 threshold range(0，max)
", "rawMarkdown": "i train my model using d1dense and val on d3dense, local surfacedice is 0.857,but lb is only 0.84, and the public 0.859, i test on d3 dense with local surface dice only 0.806.😥 threshold range(0，max)", "votes": null } ], "comments": [ { "id": 2620540, "author_name": "yuhanggong", "author_url": "", "post_date": "01/26/2024 07:38:12", "content": "Currently, my personal experiment process for exploring density is as follows: 1) Find the best validation dataset 2) Find the best training data. My conclusion is that dense1 and dense3 are the closest to the public test dataset. However, I’m not sure if it’s a problem with the experimental setup, so I’d like to discuss this with everyone.
i train my model using d1dense and val on d3dense, local surfacedice is 0.857,but lb is only 0.84, and the public 0.859, i test on d3 dense with local surface dice only 0.806.😥 threshold range(0，max)
", "votes": null, "replies": [] } ] }, { "id": 2620543, "author_name": "sacuscreed", "author_url": "", "post_date": "01/26/2024 07:44:24", "content": "Hi. In my opinion. If you are scoring the performance on segmentation. They should be as dense as they could.
", "votes": null, "replies": [ { "id": 2620600, "author_name": "yuhanggong", "author_url": "", "post_date": "01/26/2024 08:33:23", "content": "Thank you！It makes sense! I have carefully read the process of maual annotation https://arxiv.org/pdf/2311.13319.pdf
", "votes": null, "replies": [] } ] }, { "id": 2620597, "author_name": "yuhanggong", "author_url": "", "post_date": "01/26/2024 08:30:58", "content": "I have another question: Are the test datasets sliced along the z-axis？
", "votes": null, "replies": [ { "id": 2620734, "author_name": "sacuscreed", "author_url": "", "post_date": "01/26/2024 10:26:57", "content": "https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/460833
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2620515": "This might be a straightforward question, but I’m curious to know based on your experiments, is the public test closer to densely segmented or sparsely segmented? Thank you！", "2620540": "Currently, my personal experiment process for exploring density is as follows: 1) Find the best validation dataset 2) Find the best training data. My conclusion is that dense1 and dense3 are the closest to the public test dataset. However, I’m not sure if it’s a problem with the experimental setup, so I’d like to discuss this with everyone.", "2620543": "Hi. In my opinion. If you are scoring the performance on segmentation. They should be as dense as they could.", "2620597": "I have another question: Are the test datasets sliced along the z-axis？", "2620600": "Thank you！It makes sense! I have carefully read the process of maual annotation [https://arxiv.org/pdf/2311.13319.pdf](url)", "2620734": "https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/460833", "2623124": "i train my model using d1dense and val on d3dense, local surfacedice is 0.857,but lb is only 0.84, and the public 0.859, i test on d3 dense with local surface dice only 0.806.😥 threshold range(0，max)" }, "source": "meta" }