![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F7965939%2Fe419b6ec81ee8b5b95933e611cd85054%2FID1254829053.jpg?generation=1761740392834863&alt=media\")
Just checking some samples manually and this image here has zero clover mass despite clearly having clovers in the top (slightly on the right) of the image (train/ID1254829053.jpg)
\nJust wondering if theres any available limits of detection (i.e., values < small value were reject?)
\n
Just checking some samples manually and this image here has zero clover mass despite clearly having clovers in the top (slightly on the right) of the image (train/ID1254829053.jpg)
\nJust wondering if theres any available limits of detection (i.e., values < small value were reject?)
\n
Thank you for your careful observation and findings.
\nAccording to the reference paper, the data collection method involved: mowing → mixing → sampling → sorting → weighing. Because classifying and oven-drying all the grass in the entire image would be extremely labor-intensive, a certain proportion of random sampling was applied. In this image, there are only a very small number of clovers, and they are located at the edge. They might have been blowe away during mowing, possibly not accurately cut, or unfortunately excluded during random sampling—or a combination of these factors. This should be considered a labeling error rather than a systematic limit of detection.
", "rawMarkdown": "Thank you for your careful observation and findings.\n\nAccording to the reference paper, the data collection method involved: mowing → mixing → sampling → sorting → weighing. Because classifying and oven-drying all the grass in the entire image would be extremely labor-intensive, a certain proportion of random sampling was applied. In this image, there are only a very small number of clovers, and they are located at the edge. They might have been blowe away during mowing, possibly not accurately cut, or unfortunately excluded during random sampling—or a combination of these factors. This should be considered a labeling error rather than a systematic limit of detection.", "votes": null }, { "id": "3308685", "postDate": "10/30/2025 03:00:17", "content": "Since all data comes from the same paper (including test data), will there be such problems in the private test set as well?
", "rawMarkdown": "Since all data comes from the same paper (including test data), will there be such problems in the private test set as well?", "votes": null }, { "id": "3308689", "postDate": "10/30/2025 03:26:08", "content": "I guess there will be
", "rawMarkdown": "I guess there will be", "votes": null }, { "id": "3308833", "postDate": "10/30/2025 11:12:44", "content": "Will there be any corrections done then, on the test set? Because having these problems in test set defeats the purpose. If my model predicts even smaller occurrences very accurately, it will lead to a poorer score in test set given this issue persists.
", "rawMarkdown": "Will there be any corrections done then, on the test set? Because having these problems in test set defeats the purpose. If my model predicts even smaller occurrences very accurately, it will lead to a poorer score in test set given this issue persists.", "votes": null }, { "id": "3309088", "postDate": "10/30/2025 23:19:48", "content": "Unfortunately, I believe corrections are not feasible. As described earlier, all values are from destructive sampling and irreversible post-processing, rather than being estimated by human eyes. Artificially modifying these data makes no sense and could potentially introduce systematic biases.
", "rawMarkdown": "Unfortunately, I believe corrections are not feasible. As described earlier, all values are from destructive sampling and irreversible post-processing, rather than being estimated by human eyes. Artificially modifying these data makes no sense and could potentially introduce systematic biases.", "votes": null } ], "comments": [ { "id": 3308665, "author_name": "qiyuliaocsiro", "author_url": "", "post_date": "10/29/2025 23:54:45", "content": "Thank you for your careful observation and findings.
\nAccording to the reference paper, the data collection method involved: mowing → mixing → sampling → sorting → weighing. Because classifying and oven-drying all the grass in the entire image would be extremely labor-intensive, a certain proportion of random sampling was applied. In this image, there are only a very small number of clovers, and they are located at the edge. They might have been blowe away during mowing, possibly not accurately cut, or unfortunately excluded during random sampling—or a combination of these factors. This should be considered a labeling error rather than a systematic limit of detection.
", "votes": null, "replies": [ { "id": 3308685, "author_name": "p4rallax", "author_url": "", "post_date": "10/30/2025 03:00:17", "content": "Since all data comes from the same paper (including test data), will there be such problems in the private test set as well?
", "votes": null, "replies": [ { "id": 3308689, "author_name": "qiyuliaocsiro", "author_url": "", "post_date": "10/30/2025 03:26:08", "content": "I guess there will be
", "votes": null, "replies": [ { "id": 3308833, "author_name": "p4rallax", "author_url": "", "post_date": "10/30/2025 11:12:44", "content": "Will there be any corrections done then, on the test set? Because having these problems in test set defeats the purpose. If my model predicts even smaller occurrences very accurately, it will lead to a poorer score in test set given this issue persists.
", "votes": null, "replies": [ { "id": 3309088, "author_name": "qiyuliaocsiro", "author_url": "", "post_date": "10/30/2025 23:19:48", "content": "Unfortunately, I believe corrections are not feasible. As described earlier, all values are from destructive sampling and irreversible post-processing, rather than being estimated by human eyes. Artificially modifying these data makes no sense and could potentially introduce systematic biases.
", "votes": null, "replies": [] } ] } ] } ] } ] } ], "raw_markdown_by_id": { "3308432": "Just checking some samples manually and this image here has zero clover mass despite clearly having clovers in the top (slightly on the right) of the image (train/ID1254829053.jpg)\n\nJust wondering if theres any available limits of detection (i.e., values < small value were reject?)\n\n", "3308665": "Thank you for your careful observation and findings.\n\nAccording to the reference paper, the data collection method involved: mowing → mixing → sampling → sorting → weighing. Because classifying and oven-drying all the grass in the entire image would be extremely labor-intensive, a certain proportion of random sampling was applied. In this image, there are only a very small number of clovers, and they are located at the edge. They might have been blowe away during mowing, possibly not accurately cut, or unfortunately excluded during random sampling—or a combination of these factors. This should be considered a labeling error rather than a systematic limit of detection.", "3308685": "Since all data comes from the same paper (including test data), will there be such problems in the private test set as well?", "3308689": "I guess there will be", "3308833": "Will there be any corrections done then, on the test set? Because having these problems in test set defeats the purpose. If my model predicts even smaller occurrences very accurately, it will lead to a poorer score in test set given this issue persists.", "3309088": "Unfortunately, I believe corrections are not feasible. As described earlier, all values are from destructive sampling and irreversible post-processing, rather than being estimated by human eyes. Artificially modifying these data makes no sense and could potentially introduce systematic biases." }, "source": "meta" }