GitLab

Summary: The default mapper may load "file_name" and "sem_seg_file_name" from `dataset_dict`, when prefetching them from manifold, we no longer need to load them because they're already fetched. This diff adds two more fields for holding those pre-fetched data, and make the mapper work in both cases.

Reviewed By: newstzpz

Differential Revision: D26972340

fbshipit-source-id: 63f6dc809d321e149aa5adf9f92c3ace07cbf2a7

Summary:
Evaluate the predictor generated by previous step.
This diff modify the lightning_train_net to reuse the evaluation logic by adding a `predictor_path` param.
This diff also makes Lightning training backend depends on `cfg.MODEL.DEVICE` so that in evaluate_predictor step, user could set backend by changing model device. This is useful for evaluating int8 quantized model.

Reviewed By: newstzpz

Differential Revision: D27150609

fbshipit-source-id: fb72da3e81db932c0fa479350150720143e09a3e

Summary: As titled.

Reviewed By: newstzpz

Differential Revision: D27074737

fbshipit-source-id: 72f2535fc730a37f5ea8f58aaff88005c28ffc5b

Summary:
Given that the way to create D2 (https://github.com/facebookresearch/d2go/commit/465cdb842513eb910aa20fcedea1d2edd15dc7b7)go runner and Lightning task are different, get_class was introduced so that in application we could do:
```
if is Lightning:
    task_cls = get_class(classname)
    task = task_cls(cfg)
else:
    runner = create_runner(classname)
```
It turns out that we could need to do that in many places: workflow, binaries.
This diff revert `get_class` and return class in `create_runner` if the class is a Lightning module.

Reviewed By: newstzpz

Differential Revision: D26676595

fbshipit-source-id: c3ce2016d09fe073af4c2dd9f98eea4e59ca621b

Summary: Prepare the launch script for IG, which support registering new dataset for GANs on the fly

Reviewed By: newstzpz

Differential Revision: D27211763

fbshipit-source-id: f79978ceae246ab4f27a8083d25dd50c62dcefab

Summary: Not d2go.tests is not a library for oss, move utils code to d2go.utils.testing

Reviewed By: zhanghang1989

Differential Revision: D26706933

fbshipit-source-id: 85767b66bbb6c67db05e11823beb4840220b2aa3

Summary:
`checkpoint_callback` is being phased out. Initially, it was a special way to configure checkpoints, but it makes more sense for those callbacks to be included in the general `callbacks` trainer argument. In 1.2.X, `checkpoint_callback` is expected to be a boolean value only.

If `checkpoint_callback=False` **and** an instance of `ModelCheckpoint` is passed in the trainer's `callbacks` arguments, Lightning raises a [misconfiguration error](https://github.com/PyTorchLightning/pytorch-lightning/blob/2f6ce1ae7fff34d16d3707571f6a9a7b0fb0c50a/pytorch_lightning/trainer/connectors/callback_connector.py#L66-L70)

Reviewed By: newstzpz

Differential Revision: D27139315

fbshipit-source-id: 07ad5ea520583a2e46a9cb2a938f98968265c932

Summary: Add option to specify a custom subclass id mapping. Allows for flexibility when training models with different outputs needed.

Reviewed By: sanjeevk42

Differential Revision: D26826986

fbshipit-source-id: 9dba4f0f2f2afebd2f152ddd9aebd46cf4c86a0d

Summary: Pull Request resolved: https://github.com/facebookresearch/d2go/pull/24

Reviewed By: wat3rBro

Differential Revision: D27127642

Pulled By: zhanghang1989

fbshipit-source-id: 18bc3c2fa05232cacc778925db6b7dcea99b108c

Summary:
As titled.
The logs will be stored in scuba_caffe2_pytorch_usage_stats and can be queried like
```
WITH events AS (
    SELECT
        DISTINCT workflow_run_id,
        REGEXP_EXTRACT(event, 'D2 (https://github.com/facebookresearch/d2go/commit/465cdb842513eb910aa20fcedea1d2edd15dc7b7)Go.Runner\.([a-zA-Z0-9_]*)', 1)
            AS runner_name,
        ds
    FROM scuba_caffe2_pytorch_usage_stats
    WHERE
        ds BETWEEN '$START_DATE$' AND '$END_DATE$'
        AND event LIKE '%D2 (https://github.com/facebookresearch/d2go/commit/465cdb842513eb910aa20fcedea1d2edd15dc7b7)Go.Runner%'
        AND flow_is_test = 0
        AND flow_is_local_run = 0
        AND workflow_run_id > 0

)
SELECT
    COUNT(1) AS total_runs,
    runner_name,
    ds
FROM events
GROUP BY
    ds, runner_name
ORDER BY
    total_runs DESC
```

Reviewed By: colin2328

Differential Revision: D26032225

fbshipit-source-id: ab1e06f3b1af200baf530506be9b3894ddf77126

Summary: Extend conversion to support ids beyond cocotext format where ids are strings.

Reviewed By: newstzpz

Differential Revision: D27018211

fbshipit-source-id: 7282fd4b9a7e9cd19323235ed1a3c3e7b33cb6b4

Reviewed By: sanjeevk42

Differential Revision: D26922320

fbshipit-source-id: 8d6b4bc709d931020483284febede8cf8904d90f

Summary:
# context

Right now, when we register Person or Hair data we have a mask directory with .PNGs inside and when we register MCS data we have multiple directories <class>_mask. These are 2 separate file structures to maintain.

# this diff

**This diff is to add support for the following file structure for *both* Multi-Class and Single-Class Semantic Segmentation:**

```
mask/
  person/
  hair/
  skin/
  beard/
  clothes/
```

**without breaking the previous format for Single-Class Semantic Segmentation (`mask/` with .PNGs).**

# why?

With this change, we'll be able to prepare data once (create the file structure and d2go coco jsons) and make it ready-to-be-registered for Single-Class Semantic Segmentation (Person or Hair) and Multi-Class Segmentation.

We'll save on having to prepare the dataset 3 times and managing multiple copies of the same images and annotations. There will be one source of truth. For UGC, we'll be able to enforce privacy deletions with less complications.

Also, this is the format used in the Video Eval tool. So data prepared for training can be used for video evaluation if needed and data prepared for video evaluation can easily be prepared for training.

# to dos:

- ~~Add the v2 path support on the Single-Class Segmentation's `SemSegEvaluator` so that the v2 format can be used for evaluation (sync with wat3rBro for best way to do this)~~
  - **Done.** This happens automatically given the the Single-Class case in the `MultiSemSegEvaluator` class doesn't re-register the datasets. See the first job for an example.

Reviewed By: wat3rBro

Differential Revision: D26795469

fbshipit-source-id: 6ebbb1a430c049dad361d56842a587c76926ee23

Summary:
**Added a Issue template**

Issue : https://github.com/facebookresearch/d2go/issues/8

Pull Request resolved: https://github.com/facebookresearch/d2go/pull/15

Reviewed By: newstzpz

Differential Revision: D26933233

Pulled By: zhanghang1989

fbshipit-source-id: af4af309fae3432021f15a6aecbbfb62e0c58be2

Summary:
Pull Request resolved: https://github.com/facebookresearch/d2go/pull/17

Use PyTorch Lightning checkpoint in the test.

Reviewed By: zhanghang1989

Differential Revision: D26962697

fbshipit-source-id: abe635e374c3ada130243f0eaadff34204f04fa1

Summary:
Updated broken link to quantization-aware training section

Pull Request resolved: https://github.com/facebookresearch/d2go/pull/14

Reviewed By: newstzpz

Differential Revision: D26876105

Pulled By: zhanghang1989

fbshipit-source-id: b00bd08da2c8c8e478026ac154a2ca83fd4b19c9

Summary:
Added model fusing for fp32 models for export.
* We should fuse the fp32 model as well.

Reviewed By: wat3rBro

Differential Revision: D26785487

fbshipit-source-id: 6c14f746fd9eeb307b8ae465edbd4ef1335c9dd1

Summary:
Skipped broken unit tests for exporter.
* The qat exporter test is not working. See T85959661

Reviewed By: zhanghang1989

Differential Revision: D26785458

fbshipit-source-id: 813316e110698eeeb0b3afdf93d97b30f0b86508

Reviewed By: newstzpz

Differential Revision: D26072333

fbshipit-source-id: 6727b34458d410e904045aa58f81c3e09111882a

Summary: PathManager is the old version, therefore config files are not saved to Manifold

Reviewed By: sstsai-adl

Differential Revision: D26882435

fbshipit-source-id: 089c53db90080737b10bafe43d154d34221a7ce2

Summary:
fixes https://github.com/facebookresearch/d2go/issues/9

Pull Request resolved: https://github.com/facebookresearch/d2go/pull/13

Reviewed By: wat3rBro

Differential Revision: D26870048

Pulled By: zhanghang1989

fbshipit-source-id: 29298bca7a59aad214976aaa37461e3d316132d8

Reviewed By: zhanghang1989

Differential Revision: D26867453

fbshipit-source-id: 4c1c0ae8fcf0be5508f50f22d604d07350daabe1

Summary: Update the latency number in d2go

Reviewed By: newstzpz

Differential Revision: D26817844

fbshipit-source-id: 6a2e73956201ca58efafef0fcdef7a5f06ffea7b

Summary: migrated to iopath for d2go - rest.

Reviewed By: zhanghang1989

Differential Revision: D26632160

fbshipit-source-id: 182fcd9ad493a483ab7f342f373a7be2de7d10ca

Summary: migrated to iopath for d2go - runner.

Reviewed By: zhanghang1989

Differential Revision: D26632151

fbshipit-source-id: ff240967cf9f7ab8a68caee240ba56edaf507a02

Summary: migrated to iopath for d2go - data.

Reviewed By: sstsai-adl

Differential Revision: D26632143

fbshipit-source-id: d85e4a00817115a8762e4444406d867070d4fd0c

Summary:
Change depoyment to deployment in README.md.
Change datasest to datasets in tools/exporter.py.

Pull Request resolved: https://github.com/facebookresearch/d2go/pull/7

Reviewed By: newstzpz

Differential Revision: D26821039

Pulled By: zhanghang1989

fbshipit-source-id: 5056d15c877c4b3d771d33267139e73f1527da21

Summary:
Hi,

This PR changes `mdoel_zoo` to `model_zoo` in demo/README.md.

Pull Request resolved: https://github.com/facebookresearch/d2go/pull/6

Reviewed By: newstzpz

Differential Revision: D26821013

Pulled By: zhanghang1989

fbshipit-source-id: d7512d350daba57a8437bc269adbbf18e95a2d4c

Summary:
Use EMA weights when exporting the model. Many previous FBNetV3_B* + EMA training runs didn't have this config flag set, so setting it here during export when the model is built.

If a model uses EMA (MODEL_EMA.ENABLED = True), then this flag will take effect and use EMA weights. Otherwise, this flag has no effect on the export.

Yanghan may be adding `torchscript_mobile` export type, deprecating this script, soon.

Reviewed By: vivekn, sanjeevk42

Differential Revision: D26412081

fbshipit-source-id: 98f5402279ab02157e5e3dbefd41840a5e230d94

Summary: Pull Request resolved: https://github.com/facebookresearch/d2go/pull/5

Reviewed By: wat3rBro

Differential Revision: D26780956

Pulled By: zhanghang1989

fbshipit-source-id: 26af80bbdf6bcb6af4a8b5d27e655826b34db26a

Reviewed By: theschnitz

Differential Revision: D26230057

fbshipit-source-id: 57eb15b5eea27f65035b2509e920ba269f7999f1

Summary: As in title.

Reviewed By: SuperIRabbit

Differential Revision: D25349762

fbshipit-source-id: 69e21cc00a9dd5b70a479fd4383742593a62997e

Summary:
As titled. The OSS version only use PyTorch Lightning while internal version leverages some features(e.g. Manifold integration, every_n_step checkpointing).
This diff splits train_net.main into smaller functions so that they could be shared across OSS and internal versions.

Reviewed By: zhanghang1989

Differential Revision: D26752701

fbshipit-source-id: 7f68e2a81e78193e117517a0ff668ab14b76ea65

Summary: As titled. Make a copy of quantization callback to unblock D2go OSS.

Reviewed By: zhanghang1989

Differential Revision: D26735525

fbshipit-source-id: 12b77f04cfa1361e856b26ea218a262da1fadd88

fbshipit-source-id: f4a8ba78691d8cf46e003ef0bd2e95f170932778