Some folks climbed so fast to that range and then made marginal improvements. I can still improve my CV score to 0.38x but it doesn't translate to LB score anymore. What could be the reason?
", "messages": [ { "id": 2477104, "postDate": "2023-10-11T05:37:42.773Z", "content": "Some folks climbed so fast to that range and then made marginal improvements. I can still improve my CV score to 0.38x but it doesn't translate to LB score anymore. What could be the reason?
", "rawMarkdown": "Some folks climbed so fast to that range and then made marginal improvements. I can still improve my CV score to 0.38x but it doesn't translate to LB score anymore. What could be the reason?", "votes": 6 }, { "id": 2477597, "postDate": "2023-10-11T12:28:46.937Z", "content": "Let's welcome shake up.
", "rawMarkdown": "Let's welcome shake up.", "votes": 4, "replies": [ { "id": 2478668, "postDate": "2023-10-12T06:27:27.917Z", "content": "I think it will be on a small scale. CV/LB was correlated so far but there are so many teams stuck between 0.44 and 0.46 so all of them have chance for gold medal.
", "rawMarkdown": "I think it will be on a small scale. CV/LB was correlated so far but there are so many teams stuck between 0.44 and 0.46 so all of them have chance for gold medal." } ] }, { "id": 2477412, "postDate": "2023-10-11T08:52:26.423Z", "content": "external data, my guess.
", "rawMarkdown": "external data, my guess.", "replies": [ { "id": 2477520, "postDate": "2023-10-11T10:52:26.017Z", "content": "I am not using any external data
", "rawMarkdown": "I am not using any external data", "votes": 1, "replies": [ { "id": 2479904, "postDate": "2023-10-13T01:47:39.970Z", "content": "congrats! insight study.
\nnew algorithm or multiple fusion?
", "rawMarkdown": "congrats! insight study. \n\nnew algorithm or multiple fusion?" } ] }, { "id": 2477529, "postDate": "2023-10-11T11:02:05.213Z", "content": "I don't think there is external data for this problem. You can do indomain pretraining with similar datasets but it probably doesn't worth it.
", "rawMarkdown": "I don't think there is external data for this problem. You can do indomain pretraining with similar datasets but it probably doesn't worth it.", "replies": [ { "id": 2477640, "postDate": "2023-10-11T13:11:52.490Z", "content": "what about the dataset used to train the TotalSegmentator? https://zenodo.org/record/8367088 maybe useful for the segmenation step?
", "rawMarkdown": "what about the dataset used to train the TotalSegmentator? https://zenodo.org/record/8367088 maybe useful for the segmenation step?" }, { "id": 2477657, "postDate": "2023-10-11T13:24:30.423Z", "content": "I would be amazed if someone pulled off an effective pretraining on a 3D dataset like this for classification.
", "rawMarkdown": "I would be amazed if someone pulled off an effective pretraining on a 3D dataset like this for classification." } ] } ] }, { "id": 2477807, "postDate": "2023-10-11T15:18:11.343Z", "rawMarkdown": "", "votes": 1, "isDeleted": true, "replies": [ { "id": 2477811, "postDate": "2023-10-11T15:20:53.623Z", "content": "He is medical imaging competitions legend, probably improving scores locally throughout the competition :)
", "rawMarkdown": "He is medical imaging competitions legend, probably improving scores locally throughout the competition :) ", "votes": 2, "replies": [ { "id": 2478537, "postDate": "2023-10-12T04:36:13.533Z", "content": "Indeed - see his helpful post Active Extravasation Bounding Boxes
\nThink domain knowledge and ability to do effective data preprocessing and filtering for best training outcomes helps.
Let's welcome shake up.
", "votes": 4, "replies": [ { "id": 2478668, "author_name": "Gunes Evitan", "author_url": "", "post_date": "2023-10-12T06:27:27.917000", "content": "I think it will be on a small scale. CV/LB was correlated so far but there are so many teams stuck between 0.44 and 0.46 so all of them have chance for gold medal.
", "votes": 0, "replies": [] } ] }, { "id": 2477412, "author_name": "dragon zhang", "author_url": "", "post_date": "2023-10-11T08:52:26.423000", "content": "external data, my guess.
", "votes": 0, "replies": [ { "id": 2477520, "author_name": "Theo Viel", "author_url": "", "post_date": "2023-10-11T10:52:26.017000", "content": "I am not using any external data
", "votes": 1, "replies": [ { "id": 2479904, "author_name": "dragon zhang", "author_url": "", "post_date": "2023-10-13T01:47:39.970000", "content": "congrats! insight study.
\nnew algorithm or multiple fusion?
", "votes": 0, "replies": [] } ] }, { "id": 2477529, "author_name": "Gunes Evitan", "author_url": "", "post_date": "2023-10-11T11:02:05.213000", "content": "I don't think there is external data for this problem. You can do indomain pretraining with similar datasets but it probably doesn't worth it.
", "votes": 0, "replies": [ { "id": 2477640, "author_name": "Johnny Hyland", "author_url": "", "post_date": "2023-10-11T13:11:52.490000", "content": "what about the dataset used to train the TotalSegmentator? https://zenodo.org/record/8367088 maybe useful for the segmenation step?
", "votes": 0, "replies": [] }, { "id": 2477657, "author_name": "Gunes Evitan", "author_url": "", "post_date": "2023-10-11T13:24:30.423000", "content": "I would be amazed if someone pulled off an effective pretraining on a 3D dataset like this for classification.
", "votes": 0, "replies": [] } ] } ] }, { "id": 2477807, "author_name": "", "author_url": "", "post_date": "2023-10-11T15:18:11.343000", "content": "", "votes": 1, "replies": [ { "id": 2477811, "author_name": "Nischay Dhankhar", "author_url": "", "post_date": "2023-10-11T15:20:53.623000", "content": "He is medical imaging competitions legend, probably improving scores locally throughout the competition :)
", "votes": 2, "replies": [ { "id": 2478537, "author_name": "something4kag", "author_url": "", "post_date": "2023-10-12T04:36:13.533000", "content": "Indeed - see his helpful post Active Extravasation Bounding Boxes
\nThink domain knowledge and ability to do effective data preprocessing and filtering for best training outcomes helps.