{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":510796129,"defaultBranch":"main","name":"Knowledge-Distillation-in-Semantic-Segmentation","ownerLogin":"JakubTomaszewski","currentUserCanPush":false,"isFork":false,"isEmpty":false,"createdAt":"2022-07-05T15:35:58.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/57564656?v=4","public":true,"private":false,"isOrgOwned":false},"refInfo":{"name":"","listCacheKey":"v0:1681944498.0","currentOid":""},"activityList":{"items":[{"before":"113d5c6eb1334241fa287c50c65bc1bbcca3a91d","after":"fe1014b8ed39b20d5a552e1edced7d4f764d5ebb","ref":"refs/heads/main","pushedAt":"2023-11-14T10:01:24.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Docs: Add images to README","shortMessageHtmlLink":"Docs: Add images to README"}},{"before":"9d8d54c974bd30541983043628a70fb9a342bec6","after":"113d5c6eb1334241fa287c50c65bc1bbcca3a91d","ref":"refs/heads/main","pushedAt":"2023-06-19T21:51:37.690Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Remove unnecessary feature-based KD losses","shortMessageHtmlLink":"Remove unnecessary feature-based KD losses"}},{"before":"9bb270dd542fe9c045999c6fe63b00043f617515","after":"9d8d54c974bd30541983043628a70fb9a342bec6","ref":"refs/heads/main","pushedAt":"2023-06-19T21:45:35.022Z","pushType":"push","commitsCount":2,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Docs: Add docstring to scripts\n\nAdd docstrings describing the usage of provided scripts.","shortMessageHtmlLink":"Docs: Add docstring to scripts"}},{"before":"544fea576cf1a6634a44da7cddcee381d8677de9","after":"9bb270dd542fe9c045999c6fe63b00043f617515","ref":"refs/heads/main","pushedAt":"2023-06-12T17:59:28.614Z","pushType":"push","commitsCount":29,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Feat: Add class representing a model ensemble\n\nThe class has been used for evaluating the performance of a model ensemble","shortMessageHtmlLink":"Feat: Add class representing a model ensemble"}},{"before":"b73b57fad3f2791d29d68bf5247b1eafba406a76","after":"c2e7eebd77b43f741d7fd66c17e694758279753a","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-05-01T17:03:28.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Add normalizing feature loss by dividing it by the number of network layers.","shortMessageHtmlLink":"Add normalizing feature loss by dividing it by the number of network …"}},{"before":"7dd648b94641b0b7c457ed26e76d96aabff49a6b","after":"b73b57fad3f2791d29d68bf5247b1eafba406a76","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-05-01T10:26:33.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Add temperature and normalization to the `FeatureMapDistillationKLDivLoss`","shortMessageHtmlLink":"Add temperature and normalization to the `FeatureMapDistillationKLDiv…"}},{"before":"ed999ac8239f3fd48b809b5bdf2560d2eb33297c","after":"7dd648b94641b0b7c457ed26e76d96aabff49a6b","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-05-01T03:16:26.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Change the Feature-based loss to the KLDiv Loss\n\nChange the student model from SegFormer B0 to SegFormer B1","shortMessageHtmlLink":"Change the Feature-based loss to the KLDiv Loss"}},{"before":"79940dba2d655a071ae72582d4098af13d79bb9a","after":"ed999ac8239f3fd48b809b5bdf2560d2eb33297c","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-04-29T22:08:07.000Z","pushType":"push","commitsCount":3,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Change the implementation of `FeatureMapDistillationMSELoss`\n\nPreviously the mismatching student layers were interpolated to match the size of the teacher layer. Conducted experiments show, that this method does not improve the model's performance (IoU stuck at 63% vs vanilla B0 achieved 65.5%).\nTherefore, the implementation has been changed to calculate the MSE only on the matching number of channels.","shortMessageHtmlLink":"Change the implementation of <code>FeatureMapDistillationMSELoss</code>"}},{"before":"128e769c3445f04b25193ebe905d29a073203633","after":"79940dba2d655a071ae72582d4098af13d79bb9a","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-04-27T10:20:48.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Fix: Fix missing teacher model transfer to compute device","shortMessageHtmlLink":"Fix: Fix missing teacher model transfer to compute device"}},{"before":"0241bcf2a6f96c4275caf57c09ad64a95b30d46d","after":"128e769c3445f04b25193ebe905d29a073203633","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-04-26T23:57:41.000Z","pushType":"push","commitsCount":3,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Feat: Add training script for performing feature knowledge distillation\n\nCreate a new training script for performing feature knowledge.\nFurthermore, created a new config for this task in the `config.py` file.","shortMessageHtmlLink":"Feat: Add training script for performing feature knowledge distillation"}},{"before":"891f4075250d89261c4bbc1e22c08af91c627efe","after":"0241bcf2a6f96c4275caf57c09ad64a95b30d46d","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-04-22T22:20:17.000Z","pushType":"push","commitsCount":3,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Remove proportional weighting the hard labels loss","shortMessageHtmlLink":"Remove proportional weighting the hard labels loss"}},{"before":"3cca503962d7085529018261d2f0838a06421f00","after":"891f4075250d89261c4bbc1e22c08af91c627efe","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-04-22T10:57:52.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Config: KD Batch size change and KD Loss change.\n\n Batch size decreased from 4 -> 2 to fit on the GPU\n\nLoss func changed from CrossEntropy -> KLDiv according to the approaches presented in most papers","shortMessageHtmlLink":"Config: KD Batch size change and KD Loss change."}},{"before":"10240f0365c88828b383b664824a5047ceb90a75","after":"3cca503962d7085529018261d2f0838a06421f00","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-04-20T09:58:18.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Fix: Remove redundant model output in KD Trainer class\n\nRemove redundant model output in KD Trainer class, which caused the code to crash in the evaluation loop.","shortMessageHtmlLink":"Fix: Remove redundant model output in KD Trainer class"}},{"before":"259bc28e45c83519a664bc292ff34e4ff90220ef","after":"10240f0365c88828b383b664824a5047ceb90a75","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-04-20T00:08:50.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Fix missing model transfer to device in KD trainer\n\nThere was a missing `teacher.to(device)` in the KD trainer, which caused the teacher model to be on CPU, whereas the student could be on GPU. This caused a crash when the teacher model was used to compute the loss.","shortMessageHtmlLink":"Fix missing model transfer to device in KD trainer"}},{"before":"8eb684049d3dca0da7ea373c64f82866583f2d5f","after":"259bc28e45c83519a664bc292ff34e4ff90220ef","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-04-19T23:55:49.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Fix incorrect config parser","shortMessageHtmlLink":"Fix incorrect config parser"}},{"before":"fd17a6d0dd2052f7fdc39ee2c0ba2825b0711085","after":"544fea576cf1a6634a44da7cddcee381d8677de9","ref":"refs/heads/main","pushedAt":"2023-04-19T23:52:24.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Fix incorrect config parser\n\nThe incorrect config parser caused some arguments to be missing","shortMessageHtmlLink":"Fix incorrect config parser"}},{"before":"4c13802e50288cbe4b0a2b5d60de59af9d7f4c6c","after":"8eb684049d3dca0da7ea373c64f82866583f2d5f","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-04-19T23:19:36.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Add script for training the model with knowledge distillation","shortMessageHtmlLink":"Add script for training the model with knowledge distillation"}},{"before":"c0e6d86e86e031a68b06f27b96485aa07c1084df","after":"fd17a6d0dd2052f7fdc39ee2c0ba2825b0711085","ref":"refs/heads/main","pushedAt":"2023-04-19T23:16:56.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Change output evaluation path","shortMessageHtmlLink":"Change output evaluation path"}},{"before":null,"after":"4c13802e50288cbe4b0a2b5d60de59af9d7f4c6c","ref":"refs/heads/knowledge-distillation","pushedAt":"2023-04-19T22:48:18.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Add script for training the model with knowledge distillation","shortMessageHtmlLink":"Add script for training the model with knowledge distillation"}},{"before":"d735b09ce268d6de269d3c879f95c893a388c051","after":"c0e6d86e86e031a68b06f27b96485aa07c1084df","ref":"refs/heads/main","pushedAt":"2023-04-19T15:24:14.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Redundant `break` statement","shortMessageHtmlLink":"Redundant <code>break</code> statement"}},{"before":"0b29e876a9b151d529e85dcce2e06e4bc40414fc","after":"d735b09ce268d6de269d3c879f95c893a388c051","ref":"refs/heads/main","pushedAt":"2023-04-19T15:22:41.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Change output eval iou score path","shortMessageHtmlLink":"Change output eval iou score path"}},{"before":"8be7c80401d61f3800df781c3f87b717757c4ab8","after":"0b29e876a9b151d529e85dcce2e06e4bc40414fc","ref":"refs/heads/main","pushedAt":"2023-04-19T15:16:27.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Remove incorrect `break` statement","shortMessageHtmlLink":"Remove incorrect <code>break</code> statement"}},{"before":"5c77cc27d14af0338cf9dafb565495bad1cec953","after":"8be7c80401d61f3800df781c3f87b717757c4ab8","ref":"refs/heads/main","pushedAt":"2023-04-19T15:15:31.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Add saving IoU score results to json file","shortMessageHtmlLink":"Add saving IoU score results to json file"}},{"before":"8d700aee84a6812842eef05e972426d552864ce6","after":"5c77cc27d14af0338cf9dafb565495bad1cec953","ref":"refs/heads/main","pushedAt":"2023-04-19T14:42:43.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Change iou plot color from green to red to match the format of other plots","shortMessageHtmlLink":"Change iou plot color from green to red to match the format of other …"}},{"before":"adfb0be50eff120abe84db670423713e310eb321","after":"8d700aee84a6812842eef05e972426d552864ce6","ref":"refs/heads/main","pushedAt":"2023-04-19T13:57:32.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Add missing plotting functions for generating train report","shortMessageHtmlLink":"Add missing plotting functions for generating train report"}},{"before":"3be8375c8c9e2ba17c10ec2393d3725462dcd723","after":"adfb0be50eff120abe84db670423713e310eb321","ref":"refs/heads/main","pushedAt":"2023-04-19T13:52:51.000Z","pushType":"push","commitsCount":4,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Add saving the produced IoU score fig","shortMessageHtmlLink":"Add saving the produced IoU score fig"}},{"before":"545f47a84163d13c4bb7fc8442a25f057e11f652","after":"3be8375c8c9e2ba17c10ec2393d3725462dcd723","ref":"refs/heads/main","pushedAt":"2023-04-15T16:38:06.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Fix: Fix incorrect argument passing in `compute_metrics`\n\nThe order of the arguments was mismatched. This commit fixes the problem by directly getting the `EvalPrediction` value fields.","shortMessageHtmlLink":"Fix: Fix incorrect argument passing in <code>compute_metrics</code>"}},{"before":"2cdc6cd8658242366b6a83a26d106e79a2f894a6","after":"545f47a84163d13c4bb7fc8442a25f057e11f652","ref":"refs/heads/main","pushedAt":"2023-04-11T16:52:57.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Config: Change training params to fit the data on GPU\n\nDecrease batch_size and eval_accumulation_steps","shortMessageHtmlLink":"Config: Change training params to fit the data on GPU"}},{"before":"a501ba18a64ad3d248515101f9cac3b2e310f82f","after":"2cdc6cd8658242366b6a83a26d106e79a2f894a6","ref":"refs/heads/main","pushedAt":"2023-04-11T16:49:05.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Add `tensorboard` to project requirements","shortMessageHtmlLink":"Add <code>tensorboard</code> to project requirements"}},{"before":"e506c91d9c51d6bdc6bd87f14fa7122ff9c0f816","after":"a501ba18a64ad3d248515101f9cac3b2e310f82f","ref":"refs/heads/main","pushedAt":"2023-04-11T16:28:01.000Z","pushType":"push","commitsCount":2,"pusher":{"login":"JakubTomaszewski","name":"Jakub Tomaszewski","path":"/JakubTomaszewski","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/57564656?s=80&v=4"},"commit":{"message":"Refactor: Add data transformations to visualization script","shortMessageHtmlLink":"Refactor: Add data transformations to visualization script"}}],"hasNextPage":true,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"djE6ks8AAAADrqAVuwA","startCursor":null,"endCursor":null}},"title":"Activity · JakubTomaszewski/Knowledge-Distillation-in-Semantic-Segmentation"}