- 29 Aug, 2018 1 commit
-
-
Peizhao Zhang authored
Summary: Pull Request resolved: https://github.com/facebookresearch/Detectron/pull/638 Potential fix for training stuck caused by data loader failure. Reviewed By: rbgirshick Differential Revision: D9513621 fbshipit-source-id: 123974eac83f40ef2f582a90fedea790fdc442d1
-
- 24 Aug, 2018 1 commit
-
-
Ross Girshick authored
Summary: Make data loader threads daemonic so the main process can exit when an exception is raised. Reviewed By: ashwinb, newstzpz Differential Revision: D9478059 fbshipit-source-id: 00bfaab51295b63c2f95ac1be99ae56a0ccf9d94
-
- 21 Aug, 2018 1 commit
-
-
Ross Girshick authored
Summary: In some cases, eg training from scratch with a high LR or w/o activation normalization, we saw a "Negative Areas Found" exception. The RPN proposal generation code filters proposals that have a height or width that are too small. By default we use a `min_size` of 0. In a distant other part of the code, it computes the area of the proposals in order to determine FPN scale assignment. An issue arises because the `+ 1` in the `x2 - x1 + 1` width (and height) computation is not scale invariant. It turns out that the filter based on `min_size` was performed in the scaled input coordinates while the FPN scale assignment areas are computed using the unscaled, original image size. Thus a `min_size` of 0 can result in proposals with a negative area as measured wrt the original image. Uuuugh :P. This diff addresses the issue by filtering based on the area in the unscaled input image and also using a generous (though still small) margin of 1 pixel for `min_size`. Reviewed By: KaimingHe Differential Revision: D8773216 fbshipit-source-id: b4ffbc5b6a831176b2656810edea3d3d4e52d687
-
- 17 Aug, 2018 1 commit
-
-
Ross Girshick authored
Reviewed By: ir413 Differential Revision: D9360044 fbshipit-source-id: c44c392d065a2cb53db3bbd1d0f83b8ecd80a430
-
- 08 Aug, 2018 1 commit
-
-
daquexian authored
Summary: Based on orionr's work - [x] Solve [the problem about GenerateProposals](https://github.com/facebookresearch/Detectron/pull/334#issuecomment-378764468) - [x] Use the existing [ResizeNearest](https://github.com/caffe2/caffe2/blob/master/caffe2/operators/resize_op.cc#L57) layer instead of UpsampleNearest. ResizeNearest has cpu implementation and neon optimization - [x] Make it work (with https://github.com/pytorch/pytorch/pull/7091) With this PR, FPN is supported in cooperation with https://github.com/pytorch/pytorch/pull/7091. I have verified that it works on `e2e_faster_rcnn_R-50-FPN_1x.yaml` Pull Request resolved: https://github.com/facebookresearch/Detectron/pull/372 Reviewed By: newstzpz Differential Revision: D9213242 Pulled By: rbgirshick fbshipit-source-id: 8fc7b77e6cbf08adaafd760505dd760df59bfd79
-
- 03 Aug, 2018 1 commit
-
-
Ashwin Bharambe authored
Reviewed By: rbgirshick Differential Revision: D9133758 fbshipit-source-id: a698a637b294b7f2aef4c181cb47505187be3935
-
- 02 Aug, 2018 1 commit
-
-
Ashwin Bharambe authored
Summary: AFAIK, this is what the rest of the codebase does. I don't want to see logger.debug() messages show up during normal runs. Differential Revision: D9071531 fbshipit-source-id: 4ac5ef57ee47bf958e35d723c379f695554f0ef0
-
- 31 Jul, 2018 1 commit
-
-
Yanghan Wang authored
Summary: Fix typo Reviewed By: newstzpz Differential Revision: D9065042 fbshipit-source-id: 6e5bb566813febcb21c0c26afe429e2dc7a3eefc
-
- 27 Jul, 2018 1 commit
-
-
Ross Girshick authored
Reviewed By: ir413 Differential Revision: D9014510 fbshipit-source-id: a1320b1c221546ad795a94dce254e9520a0847fd
-
- 14 Jul, 2018 1 commit
-
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D8836361 fbshipit-source-id: f7d8bad14e0d6309ae349616f2b837d7082e2d74
-
- 04 Jul, 2018 1 commit
-
-
Kaiming He authored
Summary: Add TRAIN.COPY_WEIGHTS (default False): When setting True, training will copy TRAIN.WEIGHTS to checkpoint folder and treat it as a candidate checkpoint. This is useful if we want to resume a checkpoint from another job (instead of using ImageNet pre-training). Reviewed By: rbgirshick Differential Revision: D8710711 fbshipit-source-id: 9c011227867c52eaaeb40e3c2679cff1cfccdc66
-
- 03 Jul, 2018 3 commits
-
-
Gioele Ciaparrone authored
Summary: - Added command line argument to infer_simple.py to choose a custom output image format. - Added command line argument to infer_simple.py to choose to output an image even when no object is found in it. Closes https://github.com/facebookresearch/Detectron/pull/400 Reviewed By: ir413 Differential Revision: D8705039 Pulled By: rbgirshick fbshipit-source-id: fbcd707821a58004971a88946a17f005545ecab4
-
Will Kuhn authored
Summary: Closes https://github.com/facebookresearch/Detectron/pull/515 Reviewed By: ir413 Differential Revision: D8704635 Pulled By: rbgirshick fbshipit-source-id: d68d89d31c886d00a66ef7d95909eb7ebc1db9fb
-
Will Kuhn authored
Summary: Closes https://github.com/facebookresearch/Detectron/pull/511 Reviewed By: ir413 Differential Revision: D8704693 Pulled By: rbgirshick fbshipit-source-id: c05064ab69c795f8b3263fc0b35ddb666019cd5f
-
- 28 Jun, 2018 1 commit
-
-
103yiran authored
Summary: Correct the `test_batch_permutation_op.py` and `test_spatial_narrow_as_op.py` link. Closes https://github.com/facebookresearch/Detectron/pull/527 Reviewed By: rbgirshick Differential Revision: D8677148 Pulled By: ir413 fbshipit-source-id: ce8898f48590b57ed964db113035fb4da8418668
-
- 26 Jun, 2018 1 commit
-
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D8614301 fbshipit-source-id: 384df17bebd0a0186a0867cf16a7a7dfea5646ae
-
- 22 Jun, 2018 1 commit
-
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D8541427 fbshipit-source-id: f681d75ad77ea0c7411e5c6099c6bb0c2497d571
-
- 21 Jun, 2018 1 commit
-
-
Norman Mu authored
Summary: change _DEPCRECATED_KEYS -> _DEPRECATED_KEYS Differential Revision: D8553498 fbshipit-source-id: 1431a9552efce47be3aa40718c954521481950de
-
- 12 Jun, 2018 2 commits
-
-
Kaiming He authored
Summary: Update from-scratch results in GN/README.md, given the freeze_at=0 fix. Reviewed By: ir413 Differential Revision: D8369007 fbshipit-source-id: a7df13364125dac09379c5aed647f4f9a8bbf23b
-
Kaiming He authored
Summary: Move freeze_at to config. Fix GN from-scratch scripts with freeze_at=0. Reviewed By: rbgirshick Differential Revision: D8325882 fbshipit-source-id: 9f362cb269ffd962056dfe8adc888083cff3d9b2
-
- 11 Jun, 2018 1 commit
-
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D8332717 fbshipit-source-id: a3070cd83855b5c91a3395d8bbe6c4a70774a1b9
-
- 07 Jun, 2018 1 commit
-
-
Xinlei Chen authored
Summary: Remove code that seems redundant. Reviewed By: rbgirshick Differential Revision: D8200163 fbshipit-source-id: c3096c77a1c6b33eab5dc0ceb9414f3bdbbf71e9
-
- 06 Jun, 2018 1 commit
-
-
Kaiming He authored
Summary: The arg name has been changed by this diff: D7857548 Reviewed By: rbgirshick Differential Revision: D8283673 fbshipit-source-id: 961dc3773ce7b49d67874323a6dfef738538036d
-
- 01 Jun, 2018 1 commit
-
-
shenyunhang authored
Summary: I found that the json annotations for Pascal VOC downloaded from the [URL](https://storage.googleapis.com/coco-dataset/external/PASCAL_VOC.zip) in [detectron/datasets/data/README.md](https://github.com/facebookresearch/Detectron/blob/master/detectron/datasets/data/README.md#pascal-voc-annotations-in-coco-format) are split into three sets: 'train', 'val' and 'test'. Therefor, it's unable to use 'trainval' set directly. For example, the code can't found the 'voc_2007_trainval.json' file when using the following configure: ```python TRAIN: DATASETS: ('voc_2007_trainval',) Closes https://github.com/facebookresearch/Detectron/pull/299 Differential Revision: D8202450 Pulled By: ir413 fbshipit-source-id: 8d8bc8ce406b2d2ed4af57953846d21fa1968018
-
- 26 May, 2018 1 commit
-
-
Kaiming He authored
Summary: Fix a bug of aspect ratio grouping to support training batch size > 2. Reviewed By: rbgirshick Differential Revision: D8157964 fbshipit-source-id: 28eec6246cd6daf83aef30da245fe7bf340fbbc9
-
- 25 May, 2018 1 commit
-
-
Roy authored
Summary: Dilation value `cfg.MRCNN.DILATION` is only used to calculate padding for conv, but not assigned to the dilation argument of conv. Closes https://github.com/facebookresearch/Detectron/pull/388 Reviewed By: rbgirshick Differential Revision: D8160945 Pulled By: ir413 fbshipit-source-id: dca6a5cf7659f29a5d1d91e860c9dfff4963e7c3
-
- 24 May, 2018 5 commits
-
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D8117250 fbshipit-source-id: 1cd6e033c3ed3057afeeb92ed86a20d807549b5a
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D8117130 fbshipit-source-id: e75dd34c9748a17432fffe792f7a0871e09e2797
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D8116895 fbshipit-source-id: 6d8b5be01189db16ec92c8b5b0010db810777170
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D8113057 fbshipit-source-id: 36ea85bb0e0f2a9f253c96cca5fd7f148b6746ab
-
Ross Girshick authored
Reviewed By: ir413 Differential Revision: D8083267 fbshipit-source-id: 96c17d79d117bf434a5b879a8bca4c88ea6b68e2
-
- 14 May, 2018 1 commit
-
-
shenyunhang authored
Summary: We fix some typos: lib -> detectron . Closes https://github.com/facebookresearch/Detectron/pull/429 Reviewed By: rbgirshick Differential Revision: D7988278 Pulled By: ir413 fbshipit-source-id: 5e673b51a831f95aa7d7ea2ce51c5c41a1715825
-
- 09 May, 2018 1 commit
-
-
Jing Huang authored
Summary: For RoIFeatureTransform Methods with ArgMax (RoIPoolF/RoIPoolLateQuantization), the current pipeline will report ``` TypeError: I/O blob #0 of unsupported type: (BlobReference("gpu_0/roi_features"), BlobReference("gpu_0/_argmax_roi_features")) of type <type 'tuple'> ``` as in the example training: https://our.intern.facebook.com/intern/fblearner/details/51149202/ Reviewed By: viswanathgs Differential Revision: D7907518 fbshipit-source-id: 38caa51c955f5a577d11c04688b9061cb6dd3c22
-
- 08 May, 2018 2 commits
-
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D7774100 fbshipit-source-id: e11d30fd817c8553ac6f2de16924eea3f78c8814
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D7516523 fbshipit-source-id: 7c67f82c7ce6f79a66b0bdea0770d2d00735e38f
-
- 28 Apr, 2018 1 commit
-
-
Gioele Ciaparrone authored
Summary: Fixed "cannot find cmake/Cuda.cmake" error when executing "make ops" command (the correct path is cmake/legacy/Cuda.cmake") Closes https://github.com/facebookresearch/Detectron/pull/399 Reviewed By: rbgirshick Differential Revision: D7790070 Pulled By: ir413 fbshipit-source-id: 6fc49b226be227dc8e9e638a9f4c39ef2efc22aa
-
- 27 Apr, 2018 1 commit
-
-
Ilija Radosavovic authored
Reviewed By: rbgirshick Differential Revision: D7513510 fbshipit-source-id: cf46a61f8e1e9f81154fc43974be5abc4afec2f4
-
- 26 Apr, 2018 1 commit
-
-
Yangqing Jia authored
Summary: This basically allows Detectron to be built without having to include the many cmake files from Caffe2. Caffe2 now properly exports utility functions and dependent libraries, and so one can use the new style targets "caffe2_library" and "caffe2_gpu_library" directly. This should also solve problems like #251 after we update the conda version. cc pjh5. Closes https://github.com/facebookresearch/Detectron/pull/286 Reviewed By: ir413 Differential Revision: D7397696 Pulled By: Yangqing fbshipit-source-id: 7a47e0addedd4b3c88dd8e2e130afdad04706acc
-
- 25 Apr, 2018 2 commits
-
-
Kaiming He authored
Summary: fix a typo in url Reviewed By: ppwwyyxx Differential Revision: D7746833 fbshipit-source-id: a0873424ef91625d67f3fbfe44d2b1c5c9c8a4b8
-
Kaiming He authored
Summary: Add GN documentation. Reviewed By: rbgirshick, ppwwyyxx Differential Revision: D7745878 fbshipit-source-id: 8a49160b1e4026fbcf41082df4a3a8d0f8e90d85
-