{ "id": 295218, "title": "How is bad performance model by human eyes scores much better on LB", "url": "/competitions/tensorflow-great-barrier-reef/discussion/295218", "author_name": "Chenglu", "post_date": "2021-12-15T01:08:48.764000", "votes": 5, "comment_count": 6, "views": 0, "content": "

Here is the video of predictions on same local validation set: https://drive.google.com/file/d/1laeRdto5jgd4qbOGV-dLT5hVk2Y3g5GJ/view?usp=sharing

\n\n

But it's obviouse that the BLUE is much better than GREEN, what do you thing?

", "messages": [ { "id": 1618420, "postDate": "2021-12-15T01:08:48.763Z", "content": "

Here is the video of predictions on same local validation set: https://drive.google.com/file/d/1laeRdto5jgd4qbOGV-dLT5hVk2Y3g5GJ/view?usp=sharing

\n\n

But it's obviouse that the BLUE is much better than GREEN, what do you thing?

", "rawMarkdown": "Here is the video of predictions on same local validation set: https://drive.google.com/file/d/1laeRdto5jgd4qbOGV-dLT5hVk2Y3g5GJ/view?usp=sharing\n\n* **WHITE** box is ground truth.\n* **RED** box model score aroung 0.25 LB\n* **GREEN** box model score 0.45 LB\n\nBut it's obviouse that the BLUE is much better than GREEN, what do you thing?", "votes": 5 }, { "id": 1618803, "postDate": "2021-12-15T11:16:11.817Z", "content": "

I have an issue kind of something like yours, I created a discussion as well. The answer that I got, it's because of metric, it prefers recall over precision. However, in my opinion public LB's test data(%25 of all) is little poor or biased.

", "rawMarkdown": "I have an issue kind of something like yours, I created a discussion as well. The answer that I got, it's because of metric, it prefers recall over precision. However, in my opinion public LB's test data(%25 of all) is little poor or biased.", "votes": 2, "replies": [ { "id": 1619459, "postDate": "2021-12-16T00:57:12.153Z", "content": "

But the number of red box is much more than the green box in the video. Thus recall of red model should be better than green.

", "rawMarkdown": "But the number of red box is much more than the green box in the video. Thus recall of red model should be better than green. ", "votes": 3 }, { "id": 1619788, "postDate": "2021-12-16T08:04:51.263Z", "content": "

I'm totally agree with you. Apparently there is a misunderstood which I don't understand either.

", "rawMarkdown": "I'm totally agree with you. Apparently there is a misunderstood which I don't understand either.", "votes": 1 } ] }, { "id": 1618753, "postDate": "2021-12-15T09:59:57.257Z", "content": "

based on the metric.

", "rawMarkdown": "based on the metric." }, { "id": 1620879, "postDate": "2021-12-17T07:30:18.157Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 1621226, "postDate": "2021-12-17T14:45:11.217Z", "content": "

I'm pretty sure that I'm not confused with the models and colors because I was trying to figure this out for last several days.

\n

The video is for visualization, and of couse the metrics like mAP and F2 are also used.

", "rawMarkdown": "I'm pretty sure that I'm not confused with the models and colors because I was trying to figure this out for last several days.\n\nThe video is for visualization, and of couse the metrics like mAP and F2 are also used.\n" } ] } ], "comments": [ { "id": 1618803, "author_name": "Sarper Yurttas", "author_url": "", "post_date": "2021-12-15T11:16:11.817000", "content": "

I have an issue kind of something like yours, I created a discussion as well. The answer that I got, it's because of metric, it prefers recall over precision. However, in my opinion public LB's test data(%25 of all) is little poor or biased.

", "votes": 2, "replies": [ { "id": 1619459, "author_name": "Chenglu", "author_url": "", "post_date": "2021-12-16T00:57:12.153000", "content": "

But the number of red box is much more than the green box in the video. Thus recall of red model should be better than green.

", "votes": 3, "replies": [] }, { "id": 1619788, "author_name": "Sarper Yurttas", "author_url": "", "post_date": "2021-12-16T08:04:51.263000", "content": "

I'm totally agree with you. Apparently there is a misunderstood which I don't understand either.

", "votes": 1, "replies": [] } ] }, { "id": 1618753, "author_name": "dragon zhang", "author_url": "", "post_date": "2021-12-15T09:59:57.257000", "content": "

based on the metric.

", "votes": 0, "replies": [] }, { "id": 1620879, "author_name": "", "author_url": "", "post_date": "2021-12-17T07:30:18.157000", "content": "", "votes": 0, "replies": [ { "id": 1621226, "author_name": "Chenglu", "author_url": "", "post_date": "2021-12-17T14:45:11.217000", "content": "

I'm pretty sure that I'm not confused with the models and colors because I was trying to figure this out for last several days.

\n

The video is for visualization, and of couse the metrics like mAP and F2 are also used.

", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1618420": "Here is the video of predictions on same local validation set: https://drive.google.com/file/d/1laeRdto5jgd4qbOGV-dLT5hVk2Y3g5GJ/view?usp=sharing\n\n* **WHITE** box is ground truth.\n* **RED** box model score aroung 0.25 LB\n* **GREEN** box model score 0.45 LB\n\nBut it's obviouse that the BLUE is much better than GREEN, what do you thing?", "1618803": "I have an issue kind of something like yours, I created a discussion as well. The answer that I got, it's because of metric, it prefers recall over precision. However, in my opinion public LB's test data(%25 of all) is little poor or biased.", "1618753": "based on the metric.", "1620879": "" } }