GitLab

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/369

Pull Request resolved: https://github.com/facebookresearch/d2go/pull/368

Adding image subsampling in evaluation to avoid saving to TB sequential frames for video-like data.

Reviewed By: wat3rBro

Differential Revision: D39233843

fbshipit-source-id: a49bbfe1b9114c11565ed04db1f2f186675ea9ed

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/364

As title,
when `QUANTIZATION.BACKEND` is set to `turing`, call `odai.transforms` APIs instead of OSS quantization.

Also updated `image_classification` as an example to get Turing quantized model via `custom_prepare_fx/custom_convert_fx`

Reviewed By: wat3rBro

Differential Revision: D38436675

fbshipit-source-id: e4f0e02290512bce8b18c2369a67ed9b0f116825

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/367

EZ

Reviewed By: xiecong

Differential Revision: D39407416

fbshipit-source-id: d0e6fa09ff926780e98c210bfce955e6b8eec7f6

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/366

Users often time just use the default value without knowing they need to change  this config, and getting low accuracy after PTQ. Therefore increase the default value.

Reviewed By: miqueljubert

Differential Revision: D39331680

fbshipit-source-id: 6b05773c3c4cd48e6298d341d77a0e373d5439f6

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/355

switch to use inference_on_dataset_with_checkpointing in default runner.

Reviewed By: HarounH

Differential Revision: D37215292

fbshipit-source-id: c006784ce0b31700bcbb1f79c303fd791f1561ff

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/354

Allow skipping inference when running evaluation.
* `inference_on_dataset_with_checkpointing` works similar to `inference_on_dataset` in d2 but allows skipping the inference step if the evaluator has cached the results.
* If the evaluator has a function `could_skip_process` and returns True, inference will be skipped and only `evaluator. reset()` and `evaluator.evaluate()` are called.

Reviewed By: wat3rBro

Differential Revision: D37213004

fbshipit-source-id: d12cc480589ff04fd8dbb42b22633ab34bc4bf63

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/362

X-link: https://github.com/facebookresearch/detectron2/pull/4491

Recently landed D35518556 (https://github.com/facebookresearch/d2go/commit/1ffc801bf5a1d4fe926b815ba93f21632f0980f9) / Github: 36a65a0907d90ed591479b2ebaa8b61cfa0b4ef0 throws an exception with older versions of PyTorch, due to a missing library for import. This has been reported by multiple members of the PyTorch community at https://github.com/facebookresearch/detectron2/commit/36a65a0907d90ed591479b2ebaa8b61cfa0b4ef0

This change uses `try/except` to check for libraries and set flags on presence/absence to later guard code that would use them.

Reviewed By: wat3rBro

Differential Revision: D38879134

fbshipit-source-id: 72f5a7a8d350eb82be87567f006368bf207f5a74

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/358

Avoid calling scheduler.step() after the last training iteration is done

Reviewed By: wat3rBro

Differential Revision: D38605135

fbshipit-source-id: 87a55309bf6d1f7e598b567cc2372b00b8885c7c

Enable torch tracing by changing assertions in d2go forwards to allow for torch.fx.proxy.Proxy type.

Summary:
X-link: https://github.com/facebookresearch/detectron2/pull/4227

Pull Request resolved: https://github.com/facebookresearch/d2go/pull/241

Torch FX tracing propagates a type of `torch.fx.proxy.Proxy` through the graph.

Existing type assertions in the d2go code base trigger during torch FX tracing, causing tracing to fail.

This adds a check for FX tracing in progress and  adds a helper function `assert_fx_safe()`, that can be used in place of a standard assertion. This function only applies the assertion if one is not tracing, allowing d2go assertion tests to be compatible with FX tracing.

Reviewed By: wat3rBro

Differential Revision: D35518556

fbshipit-source-id: a9b5d3d580518ca74948544973ae89f8b9de3282

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/359

Currently, D2 (https://github.com/facebookresearch/d2go/commit/87374efb134e539090e0b5c476809dc35bf6aedb)Go is missing the Adam optimizer. This Diff addresses the gap.

Reviewed By: tglik, asanakoy

Differential Revision: D38492151

fbshipit-source-id: 27791c23c73942b7a466f2ca91f6b3631733ba16

Summary: Pull Request resolved: https://github.com/facebookresearch/d2go/pull/352

Reviewed By: newstzpz

Differential Revision: D37872639

fbshipit-source-id: 61acdaa669bc541dcb715af1172926efb53c0b2b

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/357

This change makes it possible to unpickle TrainNetOutput which is currently cannot be unpickled because it's a part of main module which can be different for the binary that's unpickling this dataclass.

Reviewed By: miqueljubert

Differential Revision: D38536040

fbshipit-source-id: 856594251b2eca7630d69c7917bc4746859dab9f

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/356

Attaching PDB on failure is not working when running in distributed environment. This change allows to disable this behavior by passing a command line argument.

Reviewed By: miqueljubert

Differential Revision: D38514736

fbshipit-source-id: 2e0008d6fbc6a4518a605debe67d76f8354364fc

Summary:
X-link: https://github.com/facebookresearch/detectron2/pull/4458

Pull Request resolved: https://github.com/facebookresearch/d2go/pull/353

- The QAT was using old code prior to D36786902, update to use public API
- Make `trainer:reset_data_loader` to take lazy lambda expression in order to delay the creation of dataloader. It's possible that we don't have enough RAM to hold two data loader at the same time, so we need to delete the first one, then create the second one.

Differential Revision: D38330148

fbshipit-source-id: aae28a48eabf211fe00cafe5d9ea8aeaf56e4e0c

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/349

This is to allow None, meaning model_configs is not used.

Added tasks for the other TODO.

Reviewed By: wat3rBro

Differential Revision: D38199075

fbshipit-source-id: 774ca42a82a972b7e4c642cc4306aec39e2c2f7f

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/344

we need access to the modeling hooks in EMA, e.g. build trainer.

Reviewed By: wat3rBro

Differential Revision: D37997773

fbshipit-source-id: bf4372cd310605fa35aa70f0604b084b047001d8

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/278

Allow skipping do_test after do_train.

Reviewed By: wat3rBro

Differential Revision: D36786790

fbshipit-source-id: 785556b5743ee9af2abfe6c0e9e78c7055697048

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/348

Add testcase to ensure loading from config in eval_only is covered.

Reviewed By: wat3rBro

Differential Revision: D38001319

fbshipit-source-id: e6a2edb5001ae87606a3bf48e1355037aee0f9a0

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/342

Add a cfg option to control the frequency of the writers.

Currently, the default writers include:
```
writers = [
    CommonMetricPrinter(max_iter),
    JSONWriter(os.path.join(cfg.OUTPUT_DIR, "metrics.json")),
    tbx_writer,
]
```

Reviewed By: wat3rBro

Differential Revision: D38065583

fbshipit-source-id: ebdc20aab71e03b4e18772af78b410f17ba4216d

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/347

land the fbnet architectures for video logo detection project in RCNN, FCOS and YOLO

Reviewed By: wat3rBro

Differential Revision: D38139055

fbshipit-source-id: 6ba21f482ed067c52d438e0c217e523896c2131c

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/322

TorchVision has recently added the AugMix Augmentantion. This diff adds support of the specific transform to D2 (https://github.com/facebookresearch/d2go/commit/87374efb134e539090e0b5c476809dc35bf6aedb)go

Reviewed By: newstzpz

Differential Revision: D37578243

fbshipit-source-id: b793715ccb24a3bd999a40c51d8c9a75f22110a3

Summary: Pull Request resolved: https://github.com/facebookresearch/d2go/pull/345

Reviewed By: xiecong

Differential Revision: D38086885

fbshipit-source-id: 808e104ee50c8870ae091533ac67b440e1bb8351

Summary: Pull Request resolved: https://github.com/facebookresearch/d2go/pull/343

Reviewed By: miqueljubert

Differential Revision: D38077850

fbshipit-source-id: a79541d899ce2b49a30c7f2a81a616f76321026f

Summary: Pull Request resolved: https://github.com/facebookresearch/d2go/pull/340

Reviewed By: miqueljubert

Differential Revision: D37968017

fbshipit-source-id: a3953fdbb2c48ceaffcf94df081c0b3253d247d5

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/339

add is_qat to lightning codepath

Reviewed By: jerryzh168

Differential Revision: D37937336

fbshipit-source-id: 68debe57c7f7dcf8647fad6ab9e34eff2aaa851c

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/338

Now we should've separated all the `prepare_for_quant` for eager and FX mode, we can remove this branch.

Reviewed By: jerryzh168

Differential Revision: D37865628

fbshipit-source-id: cd8f3aa7c90201f44bcfdbd65eb2edf5eded0e0c

Summary: Pull Request resolved: https://github.com/facebookresearch/d2go/pull/336

Reviewed By: jerryzh168

Differential Revision: D37860495

fbshipit-source-id: 1ce0bc7bc8071d3bfbe53cd61ed180da62e29327

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/335

Manually remove `example_input` from eager-mode-only `prepare_for_quant`.

Reviewed By: jerryzh168

Differential Revision: D37838155

fbshipit-source-id: 2d98e0264fc0c40dcf1b6f28f7fc635c52acd75e

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/333

Follow D36916149.

Reviewed By: jerryzh168

Differential Revision: D37830568

fbshipit-source-id: dbeb204ccf96dd2e90a6509f24a2864503083f60

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/332

Solve: "we have decoupled qat with `.training` in quantization, maybe we should use some flags in `cfg` instead of checking this attribute here as well"

Reviewed By: jerryzh168

Differential Revision: D37801241

fbshipit-source-id: ed9884d7b462da195ed2e07c42634acfe5beefb2

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/330

- re-enable the `test_qat`
- remove `example_input` from `DetMetaArchForTest`, since its `custom_prepare_fx` just create a tensor for avgpool.

Reviewed By: jerryzh168

Differential Revision: D37793260

fbshipit-source-id: ec7a825c61292d9c6d792f910a957c1c27832336

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/331

- remove `example_input` from `default_prepare_for_quant` since now it's eager mode only.
- rename `apply_prepare_for_quant` to `convert_to_fake_quant_model` to reflect what it's doing.

Reviewed By: jerryzh168

Differential Revision: D37794085

fbshipit-source-id: e6f12098976272d979a3aac66287d9b95432dcc8

Summary: Pull Request resolved: https://github.com/facebookresearch/d2go/pull/328

Reviewed By: jerryzh168

Differential Revision: D37695452

fbshipit-source-id: 744b1085365d1e155ea9e9fe51a6237994d90fa7

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/326

since prepare_for_quant_convert/custom_convert_fx is FX only, we can remove eager mode code from it. Also adjust the logic in `d2go/export/exporter.py` accordingly to reflect this.

Reviewed By: jerryzh168

Differential Revision: D37676977

fbshipit-source-id: ebd05082ee81bc1ac32fcc2a87bc0dfaacedd5bd

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/325

`prepare_for_quant_convert` is a confusing name because it only does `convert`, there's no "prepare" in it. It's actually for fx only, because eager mode always calls `torch.quantization.convert`, there's no way to customize it. So just call this `custom_convert_fx`. The new name is also unique in fbcode, so easy to do codemod later on.

This diff simply does the renaming by biggrep + replace.

Reviewed By: jerryzh168

Differential Revision: D37676717

fbshipit-source-id: e7d05eaafddc383dd432986267c945c8ebf94df4

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/327

Fix the bug for not leaving avgpool quantized when its input is quantized. This is not captured by CI because currently we set `small_pooler_resolution` to True to make test running faster, but this skips the `avgpool` everytime.

Reviewed By: jerryzh168

Differential Revision: D37706151

fbshipit-source-id: fde9a190324a23b97ba803e000f4caa54f6f3738

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/324

Fixes bug introduced in D37600026 (https://github.com/facebookresearch/d2go/commit/1a18ba3420e3c823accf731a11c0a91dc3babd85) -- forgot to fix imports after moving modelinghook to registry/builtin.py

Differential Revision: D37646330

fbshipit-source-id: cb763d65e7bbfd07eea6eff61727a42a6fcfbc88

Summary:
Moving MODELING_HOOK_REGISTRY to `d2go/registry/builtin.py`, so that  bootstrap works with modeling hook registry.

This fixes the detectron2 vs mobile_cv registry.

Reviewed By: wat3rBro

Differential Revision: D37600026

fbshipit-source-id: 7aa8cdec3b5eaea89d9db6fc35c7389cc9fb3193

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/321

Following up the bc-breaking change from fx graph mode quantization: https://github.com/pytorch/pytorch/pull/76496 that
added example_inputs to prepare_fx and prepare_qat_fx, we fixes the callsite related to mobile-vision and exposed
extra example_inputs in some apis

Reviewed By: wat3rBro

Differential Revision: D37163018

fbshipit-source-id: 9f0bb56659345d174a39b6d3bb4408caa553b88d

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/320

MCV/D2 (https://github.com/facebookresearch/d2go/commit/87374efb134e539090e0b5c476809dc35bf6aedb)Go's `launch` now supports `kwargs`, which matches elastic launch. Let's always use `args=(cfg, output_dir, runner_name)` for all the binaries, and use `kwargs` for remaining binary arguments (which matches the `extra_args` in FBL's OperatorArgument).

Reviewed By: sstsai-adl

Differential Revision: D37535145

fbshipit-source-id: 9767e8d71421d2262aee1fd4b9019758aa4a6bbd