…or accuracy-data curve

Summary:
Add a sampler class `RandomSubsetTrainingSampler`, which is similar to TrainingSampler but only sample a random subset (e.g., 50%) of indices.
`RandomSubsetTrainingSampler` is useful when you want to estimate the accuracy vs data-volume curves by training the model with different `subset_ratio`.

Reviewed By: ppwwyyxx

Differential Revision: D29892290

fbshipit-source-id: a342a6f1aa7852feb6566c648bd673028a3e0668

Reviewed By: ppwwyyxx

Differential Revision: D30076586

fbshipit-source-id: 72017b55bc99d6dab2b11b663cd7da98b814b544

Summary:
Pull Request resolved: facebookresearch#3322

avoid warnings like the following:
```
[W ProcessGroupNCCL.cpp:1569] Rank 0 using best-guess GPU 0 to perform barrier as devices used by
this process are currently unknown. This can potentially cause a hang if this rank to GPU mapping is
incorrect. Specify device_ids in barrier() to force use of a particular device.
```

maybe can fix the hang in facebookresearch#3319

Reviewed By: vaibhava0

Differential Revision: D30077957

fbshipit-source-id: b8827e66c5eecc06b650acde2e7ff44106327f69

Differential Revision: D26773548

fbshipit-source-id: 94bdd659b2bf35235de411795f20d6cd8f045fa4

Summary: close facebookresearch#3313

Reviewed By: vaibhava0

Differential Revision: D30062746

fbshipit-source-id: ed548c2b0513933a1071364ac4824f5346675918

Summary: close T85305354

Reviewed By: vaibhava0

Differential Revision: D30063498

fbshipit-source-id: 251a0700d92071bcb1f9eea1e8bc736119ab552a

Summary: Add cat methods to Keypoints

Reviewed By: ppwwyyxx

Differential Revision: D30199199

fbshipit-source-id: 7d2c63810e7e05931c90b7ea5102a734b7ce8af4

Reviewed By: alexander-kirillov, bowenc0221

Differential Revision: D30215015

fbshipit-source-id: 90c6a791c23f4edacdc7e7a95b41586582e8da93

Summary:
* Current code does not enable user to choose stats_mode="N" for naiveSyncBN. This change exposes that.
  - stats_mode="N" is necessary for supporting zero-length inputs. Without it:  `SyncBatchNorm(stats_mode="") does not support zero batch size.`

------------------------

Reviewed By: jonmorton, vaibhava0, ppwwyyxx

Differential Revision: D29362934

fbshipit-source-id: f8d5a33cac599f706e6620c31495e8c1577970dc

Summary:
Pull Request resolved: facebookresearch#3350

`get_local_rank` relies on a global variable set by Detectron2's `launch` utils.
Since other frameworks might use Detectron2's distribute utils but don't launch with Detectron2's `launch` utils. Use `torch.cuda.current_device` to get the current device instead.

Reviewed By: HarounH, ppwwyyxx

Differential Revision: D30233746

fbshipit-source-id: 0b140ed5c1e7cd87ccf05235127f338ffc40a53d

Summary: rank is 0-based

Reviewed By: bowenc0221

Differential Revision: D30258674

fbshipit-source-id: 1e8331e303fdcce8c96c500e520ffadc000f06ba

Summary:
avoid wanings like:
```
detectron2/structures/boxes.py:158: UserWarning: Creating a tensor from a list
of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray
with numpy.array() before converting to a tensor. (Triggered internally at torch/csrc/utils/tensor_new.cpp:201.)
```

Reviewed By: wat3rBro

Differential Revision: D30172338

fbshipit-source-id: fbe044fead72722b3c6eb9916faa2a7dca3e6eae

Summary:
The naming "boxlist" comes from maskrcnn-benchmark. This functions is created when pairwise_iou is
called `boxlist_iou`.

Rename it to follow a consistent naming.

Reviewed By: jonmorton

Differential Revision: D30428372

fbshipit-source-id: be73f5acb1c6a5a2266c2b7706394b1f84f23479

Summary:
Most production use cases needs error handling.
This is not available in pytorch's `.map()`.

This is useful as we move towards using more IterableDataset

Differential Revision: D30123656

fbshipit-source-id: 17c4caa3e4ed18c1b073bbfe8309ca3b3681a59b

Summary:
Pull Request resolved: facebookresearch#3318

Just add some typing.

fix facebookresearch#3302

Reviewed By: zhanghang1989, wat3rBro

Differential Revision: D30062501

fbshipit-source-id: c1f2a3cb42f8ee11d5e8c582fc5b8f5464c438d0

Summary: see context in T88568338

Reviewed By: wat3rBro

Differential Revision: D30063970

fbshipit-source-id: 1ad0ac0dfb72a6c7aba6e4263724550e4042c664

Summary: both used in FCOS

Reviewed By: zhanghang1989

Differential Revision: D30470063

fbshipit-source-id: 495494a669acd9f706c3845551252ced4acc9888

Summary: fix facebookresearch#3401

Reviewed By: wat3rBro

Differential Revision: D30491149

fbshipit-source-id: 0af0d850afcb4aa451da2975aa46c4c854138523

Summary: close facebookresearch#3316

Reviewed By: wat3rBro

Differential Revision: D30062593

fbshipit-source-id: e8b614c21f62012e081f425dd9df850179f1f2a7

Summary:
The error message in facebookresearch#3341
is not informative

Reviewed By: wat3rBro

Differential Revision: D30185963

fbshipit-source-id: 11fda5e6e22851942b5da9df09b6ec4e315eddef

Reviewed By: wat3rBro

Differential Revision: D29738743

fbshipit-source-id: 08b49cabe5c1c43478418600e3f0076e443d2e22

Summary: close T88689498

Reviewed By: vaibhava0

Differential Revision: D30063379

fbshipit-source-id: 23cddad66c17857f96bd60f59cf9f67291aedfff

Summary:
Implemented hook for saving best model, as discussed in facebookresearch#2114

Pull Request resolved: facebookresearch#3222

Reviewed By: zhanghang1989

Differential Revision: D29764303

Pulled By: ppwwyyxx

fbshipit-source-id: b5cb42b34a15ea151cce8fe992a9f35a964a6afe

Differential Revision: D30514760

fbshipit-source-id: f80b301ae5ea779588a9d634c1baa35f8f2d3e99

Differential Revision: D24677397

fbshipit-source-id: 1e4a991c521da1e139ccc7fe40b715dc921d3294

Summary: it has been deprecated for 1 year. moving to internal

Reviewed By: jonmorton

Differential Revision: D25876263

fbshipit-source-id: 2ee8cab3b25b66664792578bb40194d813208f2e

Summary:
fix facebookresearch#3405

The old code works for python 3.6 & 3.7; fails for python 3.9

Reviewed By: zhanghang1989

Differential Revision: D30599139

fbshipit-source-id: ab9e5d68387b68e1095112e9c2531a971567accc

Summary: run ./dev/linter.sh

Reviewed By: zhanghang1989

Differential Revision: D30599861

fbshipit-source-id: 8957443bb1ca9e143ca11a2e2675b8513ed63d16

Summary:
Thanks for your contribution!.
Fixed the following runtime Error : numpy() fails when visualizing an output element of a input batch during inference

Pull Request resolved: facebookresearch#3400

Reviewed By: jonmorton

Differential Revision: D30560943

Pulled By: ppwwyyxx

fbshipit-source-id: abe94a12b67a5c8bad2a60fff989072f99b70d16

Summary:
Pull Request resolved: facebookresearch#3410

avoid bugs like the last diff

Reviewed By: zhanghang1989

Differential Revision: D30599205

fbshipit-source-id: 3b2abd2885d384bd3f2a29b454d3da5e023e01a3

Reviewed By: zhanghang1989

Differential Revision: D30635073

fbshipit-source-id: 664ba17b768a69fed97ca94b944e43d186336966

Summary:
Wanted to use some of the latest additions to mmdetection; in particular, the UniformAssignment from YOLOF (Free box AP win, even in two stage models!).

I just cloned latest mmdetection and copied into `deeplearning/projects/third_party/mmdetection`.

Reviewed By: ppwwyyxx

Differential Revision: D30598178

fbshipit-source-id: 05fc559534a2fde41b36fc8c088c4e5c1729e7a2

Summary:
In PR facebookresearch#3311 we made densepose a pip-installable Python package. However, we forgot to include an `__init__.py` so that the `vis` module is importable. This PR adds that `__init__.py`.

It was also mentioned in that PR that there is a preference not to require torch and torchvision requirements versions:
facebookresearch#3311 (comment)

This PR also removes those requirements from the `setup.py` and lists them instead in the prerequisites in the documentation.

Pull Request resolved: facebookresearch#3382

Reviewed By: patricklabatut

Differential Revision: D30462761

Pulled By: vkhalidov

fbshipit-source-id: e58985ff24cc75ff1d7905a863fa040e3be99bc3

Summary:
## Context
- The current limit on the number of detections per image (`K`) in LVIS is 300.
- Implementing AP_pool/AP_fixed requires removing this default limit on `K`
- [Literature](https://arxiv.org/pdf/2102.01066.pdf) has shown that increasing `K` correlates with AP gains

## This Diff
- Changed limit on number of detections per image (`K`) to be customizable for LVIS and COCO through `TEST.DETECTIONS_PER_IMAGE` in the config
   - For COCO:
       - Maintain the default `max_dets_per_image` to be [1, 10, 100] as from [COCOEval](https://www.internalfb.com/code/fbsource/[88bb57c3054a]/fbcode/deeplearning/projects/cocoApi/PythonAPI/pycocotools/cocoeval.py?lines=28-29)
       - Allow users to input a custom integer for `TEST.DETECTIONS_PER_IMAGE` in the config, and use  [1, 10, `TEST.DETECTIONS_PER_IMAGE`] for COCOEval
   - For LVIS:
       - Maintain the default `max_dets_per_image` to be 300 as from [LVISEval](https://www.internalfb.com/code/fbsource/[f6b86d023721]/fbcode/deeplearning/projects/lvisApi/lvis/eval.py?lines=528-529)
       - Allow users to input a custom integer for `TEST.DETECTIONS_PER_IMAGE` in the config, and use this in LVISEval
- Added `COCOevalMaxDets` for evaluating AP with the custom limit on number of detections per image (since default `COCOeval` uses 100 as limit on detections per image for evaluating AP)

## Inference Runs using this Diff
- Performed inference using `K = {300, 1000, 10000, 100000}`
- Launched fblearner flows for object detector baseline models with N1055536 (LVIS) and N1055756 (COCO)
  - Recorded [results of running inference](https://docs.google.com/spreadsheets/d/1rgdjN2KvxcYfKCkGUC4tMw0XQJ5oZL0dwjOIh84YRg8/edit?usp=sharing)

Reviewed By: ppwwyyxx

Differential Revision: D30077359

fbshipit-source-id: 372eb5e0d7c228fb77fe23bf80d53597ec66287b

Summary: this function alone is useful in deployment

Reviewed By: zhanghang1989

Differential Revision: D30801733

fbshipit-source-id: 792f8ca016f2c6782fc25c9bbaa302588597d087

Summary:
Correct typo
to to swap -> to swap
Regarding issue facebookresearch#3458

Thanks for your contribution!

If you're sending a large PR (e.g., >100 lines),
please open an issue first about the feature / bug, and indicate how you want to contribute.

We do not always accept features.
See https://detectron2.readthedocs.io/notes/contributing.html#pull-requests about how we handle PRs.

Before submitting a PR, please run `dev/linter.sh` to lint the code.

Pull Request resolved: facebookresearch#3459

Reviewed By: zhanghang1989

Differential Revision: D30802143

Pulled By: ppwwyyxx

fbshipit-source-id: a820fa521d9e54c32d4af614ea48df3207dc6515

Reviewed By: HannaMao

Differential Revision: D30795434

fbshipit-source-id: 7a841e0f980f6c8ff40411acff9c732a53ac2c3b

Summary:
Pre-computing trimap weights avoids unnecessary CPU <--> GPU copy during
loss calculation and leads to 45% training speed up. In addition to that, this
also brings accuracy improvements. For xirp17e res 160 Portal model
pre-computing trimap weights got +0.7 IOU, +1.2 B-IOU and +0.2 BT-IOU. Note
that the Portal model still uses the trimap computation and not boundary loss.
In fact, for xirp17e boundary loss did not show additional gains compared to
trimap.

Reviewed By: wenliangzhao2018

Differential Revision: D30200036

fbshipit-source-id: 4e6c5f022b341757c93678e81d73d2e8c259b43a

Summary: Fix the keypoint evaluation pipeline.

Reviewed By: ppwwyyxx

Differential Revision: D30904414

fbshipit-source-id: ef8bff166fccf2ef059013ae97ff119f8a9b4df0

Summary: fix facebookresearch#3486

Reviewed By: zhanghang1989

Differential Revision: D30974958

fbshipit-source-id: 68c7d041f3b5b64eb0c32b4fcc99ad6d13e3542f

Summary: Make the keypoint threshold a class member, so it can be changed.

Reviewed By: ppwwyyxx

Differential Revision: D30690598

fbshipit-source-id: 29499eaf12257e5bb4871ea318dc4ab5523d304e

Summary:
Reduction of cross_entropy is always set to "mean".
Fixed to receive the reduction.

Pull Request resolved: facebookresearch#3454

Reviewed By: zhanghang1989

Differential Revision: D31052191

Pulled By: ppwwyyxx

fbshipit-source-id: 0bd992cac13c67a772fac8539ab5cc4f7c8e5c3c

Summary:
Thanks for your contribution!

If you're sending a large PR (e.g., >100 lines),
please open an issue first about the feature / bug, and indicate how you want to contribute.

We do not always accept features.
See https://detectron2.readthedocs.io/notes/contributing.html#pull-requests about how we handle PRs.

Before submitting a PR, please run `dev/linter.sh` to lint the code.

Pull Request resolved: facebookresearch#3476

Reviewed By: zhanghang1989

Differential Revision: D31052262

Pulled By: ppwwyyxx

fbshipit-source-id: d1f7ab357ce48e4d7735e024d6441a96b863bd71

Summary: Empty is a better default

Reviewed By: zhanghang1989

Differential Revision: D30980361

fbshipit-source-id: 9ad2bde6b9d9462ccf7cf953c33894a0c0279ad1

Summary:
Correct typo
standarized -> standardized
Regarding issue facebookresearch#3467

Thanks for your contribution!

If you're sending a large PR (e.g., >100 lines),
please open an issue first about the feature / bug, and indicate how you want to contribute.

We do not always accept features.
See https://detectron2.readthedocs.io/notes/contributing.html#pull-requests about how we handle PRs.

Before submitting a PR, please run `dev/linter.sh` to lint the code.

Pull Request resolved: facebookresearch#3468

Reviewed By: zhanghang1989

Differential Revision: D31052099

Pulled By: ppwwyyxx

fbshipit-source-id: 4128566943cfa54fcebd357d6c1a1f9dda70a7da

Summary: if dumped yaml is unloadable, print warning and save a pkl

Reviewed By: zhanghang1989

Differential Revision: D30975516

fbshipit-source-id: c2eb839c9ffcdee7951d78b01e5c10939debec77

Summary: it was unsupported due to omegaconf

Reviewed By: zhanghang1989

Differential Revision: D30753898

fbshipit-source-id: d47b014bdc806023ebc95364359d8695865cd3c7

Summary: Caffe2 export should be deprecated. Start from the least used model.

Reviewed By: wat3rBro

Differential Revision: D31038278

fbshipit-source-id: 18fd26942dd35f3a542f2da1e593b6c629a22ba1

Summary:
`polygons_to_bitmask` fails with empty polygon which could happen with LSJ.

Pull Request resolved: fairinternal/detectron2#564

Reviewed By: HannaMao

Differential Revision: D31181422

Pulled By: ppwwyyxx

fbshipit-source-id: 5f955e04518bae153401df0760840f5daf0da66a

Summary: Like D24677397 (facebookresearch@05bc843), but for test loader

Differential Revision: D31161853

fbshipit-source-id: 6c101843a2be681fc23b2ff241070876a77be80f