{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":372978756,"defaultBranch":"master","name":"road-anomaly-benchmark","ownerLogin":"SegmentMeIfYouCan","currentUserCanPush":false,"isFork":true,"isEmpty":false,"createdAt":"2021-06-01T22:44:12.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/84911120?v=4","public":true,"private":false,"isOrgOwned":true},"refInfo":{"name":"","listCacheKey":"v0:1691069704.0","currentOid":""},"activityList":{"items":[{"before":"b519c58f5c254da86627ddf741e034a4f97a3cd7","after":"1c7804e0749686452eb02bbcfb81234e38795f8a","ref":"refs/heads/master","pushedAt":"2023-08-24T12:48:27.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"metrics: fix best-f1 threshold selection\n\nRemove duplicate thresholds not only cmats for those duplicates","shortMessageHtmlLink":"metrics: fix best-f1 threshold selection"}},{"before":"ce850f4124255d572d1368e7291bfb48294c4e92","after":"b519c58f5c254da86627ddf741e034a4f97a3cd7","ref":"refs/heads/master","pushedAt":"2023-08-23T09:16:18.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"inference: gc","shortMessageHtmlLink":"inference: gc"}},{"before":"b580f16697b4c85345226a81942f9e8da129a9f8","after":"ce850f4124255d572d1368e7291bfb48294c4e92","ref":"refs/heads/master","pushedAt":"2023-08-10T17:34:22.000Z","pushType":"push","commitsCount":4,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"fix: np.bool -> bool","shortMessageHtmlLink":"fix: np.bool -> bool"}},{"before":"953c70d02dea384d1b26639a454174bcec607d17","after":"97ee080891108db97da2629795f57fcd30c60080","ref":"refs/heads/precision-recall-fix","pushedAt":"2023-08-10T17:31:04.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"credits\n\nThanks to Jan Ackermann from ETHZ for reporting problems with the AP curves.","shortMessageHtmlLink":"credits"}},{"before":"f5c119620e6cbc4787c04426c0a4d4192f8a39f7","after":"953c70d02dea384d1b26639a454174bcec607d17","ref":"refs/heads/precision-recall-fix","pushedAt":"2023-08-05T17:05:17.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"metrics: collapse sample counts for repeated thresholds","shortMessageHtmlLink":"metrics: collapse sample counts for repeated thresholds"}},{"before":"adbef16f85e92382b38434f4426882627ae46a55","after":"f5c119620e6cbc4787c04426c0a4d4192f8a39f7","ref":"refs/heads/precision-recall-fix","pushedAt":"2023-08-03T15:22:36.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"metrics: fixes to Precision-Recall curve\n\nThe pixel classification metrics were originally tested with smoothly distributed values. For example values output by a softmax would range from 0 to 1 and given a usual U-Net we would observe smoothly changing logit levels.\n\nThe implementation however creates wrong results in some cases - for instance if we the prediction is binary, score = 0 or 1.\n\n    In dynamic binning mode, set bin edges to surround the whole range of values by adding extra bin edges at minval - eps and maxval + eps. This allows us to correctly treat groups of sharp 0s and 1s scores.\n\n    When calculating PR curve, ensure the curve starts at precision=1,recall=0 and ends at recall=1\n\n    Area under PR curve is calculated according to sklearn and ignores the areas where the curve rises, instead taking a flat section of the graph. (see images)\n\nThere is still some mismatch when there are lots of repeated thresholds.","shortMessageHtmlLink":"metrics: fixes to Precision-Recall curve"}},{"before":"7bf85bf191d29616dae1cac02c3c484f5bf9c443","after":"b580f16697b4c85345226a81942f9e8da129a9f8","ref":"refs/heads/master","pushedAt":"2023-08-03T15:22:22.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"table: ability to load from leaderboard.json","shortMessageHtmlLink":"table: ability to load from leaderboard.json"}},{"before":"9926b49888808beea0b7230898ef45dbb4cbd4ae","after":"7bf85bf191d29616dae1cac02c3c484f5bf9c443","ref":"refs/heads/master","pushedAt":"2023-08-03T15:20:36.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"vis: expand and simplify visualization\n\nSimplify how heatmap is produced.\nExpand visualization in instance mode.","shortMessageHtmlLink":"vis: expand and simplify visualization"}},{"before":null,"after":"adbef16f85e92382b38434f4426882627ae46a55","ref":"refs/heads/precision-recall-fix","pushedAt":"2023-08-03T13:35:04.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"metrics: fixes to Precision-Recall curve\n\nThe pixel classification metrics were originally tested with smoothly distributed values. For example values output by a softmax would range from 0 to 1 and given a usual U-Net we would observe smoothly changing logit levels.\n\nThe implementation however creates wrong results in some cases - for instance if we the prediction is binary, score = 0 or 1.\n\n    In dynamic binning mode, set bin edges to surround the whole range of values by adding extra bin edges at minval - eps and maxval + eps. This allows us to correctly treat groups of sharp 0s and 1s scores.\n\n    When calculating PR curve, ensure the curve starts at precision=1,recall=0 and ends at recall=1\n\n    Area under PR curve is calculated according to sklearn and ignores the areas where the curve rises, instead taking a flat section of the graph. (see images)\n\nThere is still some mismatch when there are lots of repeated thresholds.","shortMessageHtmlLink":"metrics: fixes to Precision-Recall curve"}},{"before":"6e3ed836b6c55cacdeb8af57110687792fe7ee23","after":"9926b49888808beea0b7230898ef45dbb4cbd4ae","ref":"refs/heads/master","pushedAt":"2023-08-02T16:35:48.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"datasets: fix loader for SOD\n\nThe usual and anomaly categories can be described by ranges of classids.","shortMessageHtmlLink":"datasets: fix loader for SOD"}},{"before":"05552386cfe625da2e18c0f8efe158e5543a87e3","after":"6e3ed836b6c55cacdeb8af57110687792fe7ee23","ref":"refs/heads/master","pushedAt":"2023-08-02T15:57:07.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"adynathos","name":"Krzysztof Lis / Adynathos","path":"/adynathos","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/8000512?s=80&v=4"},"commit":{"message":"import external metrics","shortMessageHtmlLink":"import external metrics"}}],"hasNextPage":false,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"djE6ks8AAAADcamDlQA","startCursor":null,"endCursor":null}},"title":"Activity · SegmentMeIfYouCan/road-anomaly-benchmark"}