Someone please correct me if I'm wrong, the Average Precision is calculated across different thresholds, and then the Mean in MAP is calculated across all images?\nIn this case the confidence doesn't really matter and it is just???:
\n\n\n\n\nIn nearly all cases confidence will have no impact on scoring. It exists primarily to allow for submission boxes to be evaluated in a particular order to resolve extreme edge cases.
\n
As per the lyft competition
\n\nOR
\n\nAre we calculating Average precision across confidence levels (all answers are ranked by confidence) and then we take the Mean in MAP across the thresholds?
\n\nHelp! and Thank you very much in advance.\n@philculliton @addisonhoward @boliu0 @stevenwudi @dingfuzhou
", "messages": [ { "id": 689586, "postDate": "2019-12-07T05:33:21.853Z", "content": "Someone please correct me if I'm wrong, the Average Precision is calculated across different thresholds, and then the Mean in MAP is calculated across all images?\nIn this case the confidence doesn't really matter and it is just???:
\n\n\n\n\nIn nearly all cases confidence will have no impact on scoring. It exists primarily to allow for submission boxes to be evaluated in a particular order to resolve extreme edge cases.
\n
As per the lyft competition
\n\nOR
\n\nAre we calculating Average precision across confidence levels (all answers are ranked by confidence) and then we take the Mean in MAP across the thresholds?
\n\nHelp! and Thank you very much in advance.\n@philculliton @addisonhoward @boliu0 @stevenwudi @dingfuzhou
", "rawMarkdown": "Someone please correct me if I'm wrong, the Average Precision is calculated across different thresholds, and then the Mean in MAP is calculated across all images?\nIn this case the confidence doesn't really matter and it is just???: \n> In nearly all cases confidence will have no impact on scoring. It exists primarily to allow for submission boxes to be evaluated in a particular order to resolve extreme edge cases.\n\nAs per the [lyft competition](https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/overview/evaluation)\n\n**OR**\n\nAre we calculating Average precision across confidence levels (all answers are ranked by confidence) and then we take the Mean in MAP across the thresholds?\n\nHelp! and Thank you very much in advance.\n@philculliton @addisonhoward @boliu0 @stevenwudi @dingfuzhou ", "votes": 5 }, { "id": 693546, "postDate": "2019-12-12T13:54:08.487Z", "content": "They state what it does in the evaluation page but yes it is worthless, only used to sort the cars to check first.
", "rawMarkdown": "They state what it does in the evaluation page but yes it is worthless, only used to sort the cars to check first.", "votes": 2, "replies": [ { "id": 693654, "postDate": "2019-12-12T16:26:49.323Z", "content": "Oh sure
", "rawMarkdown": "Oh sure", "votes": 1 }, { "id": 695334, "postDate": "2019-12-15T01:00:47.667Z", "content": "Thank you.
", "rawMarkdown": "Thank you." } ] }, { "id": 693528, "postDate": "2019-12-12T13:31:28.347Z", "content": "It seems that probability does not influence on LB score.\nEqual submits with 0.0 and 1.0 probability values gave the same LB score
", "rawMarkdown": "It seems that probability does not influence on LB score.\nEqual submits with 0.0 and 1.0 probability values gave the same LB score", "votes": 2, "replies": [ { "id": 695335, "postDate": "2019-12-15T01:01:42.857Z", "content": "Thank you. And thanks again for the supporting explanation (concise and clear) :)
", "rawMarkdown": "Thank you. And thanks again for the supporting explanation (concise and clear) :) " } ] }, { "id": 697228, "postDate": "2019-12-17T16:28:27.953Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 693546, "author_name": "GreatGameDota", "author_url": "", "post_date": "2019-12-12T13:54:08.487000", "content": "They state what it does in the evaluation page but yes it is worthless, only used to sort the cars to check first.
", "votes": 2, "replies": [ { "id": 693654, "author_name": "Mikhail Edukov", "author_url": "", "post_date": "2019-12-12T16:26:49.323000", "content": "Oh sure
", "votes": 1, "replies": [] }, { "id": 695334, "author_name": "Rafael Natan", "author_url": "", "post_date": "2019-12-15T01:00:47.667000", "content": "Thank you.
", "votes": 0, "replies": [] } ] }, { "id": 693528, "author_name": "Mikhail Edukov", "author_url": "", "post_date": "2019-12-12T13:31:28.347000", "content": "It seems that probability does not influence on LB score.\nEqual submits with 0.0 and 1.0 probability values gave the same LB score
", "votes": 2, "replies": [ { "id": 695335, "author_name": "Rafael Natan", "author_url": "", "post_date": "2019-12-15T01:01:42.857000", "content": "Thank you. And thanks again for the supporting explanation (concise and clear) :)
", "votes": 0, "replies": [] } ] }, { "id": 697228, "author_name": "", "author_url": "", "post_date": "2019-12-17T16:28:27.953000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "689586": "Someone please correct me if I'm wrong, the Average Precision is calculated across different thresholds, and then the Mean in MAP is calculated across all images?\nIn this case the confidence doesn't really matter and it is just???: \n> In nearly all cases confidence will have no impact on scoring. It exists primarily to allow for submission boxes to be evaluated in a particular order to resolve extreme edge cases.\n\nAs per the [lyft competition](https://www.kaggle.com/c/3d-object-detection-for-autonomous-vehicles/overview/evaluation)\n\n**OR**\n\nAre we calculating Average precision across confidence levels (all answers are ranked by confidence) and then we take the Mean in MAP across the thresholds?\n\nHelp! and Thank you very much in advance.\n@philculliton @addisonhoward @boliu0 @stevenwudi @dingfuzhou ", "693546": "They state what it does in the evaluation page but yes it is worthless, only used to sort the cars to check first.", "693528": "It seems that probability does not influence on LB score.\nEqual submits with 0.0 and 1.0 probability values gave the same LB score", "697228": "" } }