ruslanchikbundyuk/VTON-API
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add at new repo again

Browse Source
This commit is contained in:
Руслан Бундюк
2025-01-28 21:48:35 +00:00
commit 6e660ddb3c
564 changed files with 75575 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

196
vton-api/preprocess/humanparsing/mhp_extension/detectron2/docs/notes/benchmarks.md Normal file
View File

@@ -0,0 +1,196 @@
# Benchmarks
Here we benchmark the training speed of a Mask R-CNN in detectron2,
with some other popular open source Mask R-CNN implementations.
### Settings
* Hardware: 8 NVIDIA V100s with NVLink.
* Software: Python 3.7, CUDA 10.1, cuDNN 7.6.5, PyTorch 1.5,
TensorFlow 1.15.0rc2, Keras 2.2.5, MxNet 1.6.0b20190820.
* Model: an end-to-end R-50-FPN Mask-RCNN model, using the same hyperparameter as the
[Detectron baseline config](https://github.com/facebookresearch/Detectron/blob/master/configs/12_2017_baselines/e2e_mask_rcnn_R-50-FPN_1x.yaml)
(it does no have scale augmentation).
* Metrics: We use the average throughput in iterations 100-500 to skip GPU warmup time.
Note that for R-CNN-style models, the throughput of a model typically changes during training, because
it depends on the predictions of the model. Therefore this metric is not directly comparable with
"train speed" in model zoo, which is the average speed of the entire training run.
### Main Results
```eval_rst
+-------------------------------+--------------------+
| Implementation | Throughput (img/s) |
+===============================+====================+
| |D2| |PT| | 62 |
+-------------------------------+--------------------+
| mmdetection_ |PT| | 53 |
+-------------------------------+--------------------+
| maskrcnn-benchmark_ |PT| | 53 |
+-------------------------------+--------------------+
| tensorpack_ |TF| | 50 |
+-------------------------------+--------------------+
| simpledet_ |mxnet| | 39 |
+-------------------------------+--------------------+
| Detectron_ |C2| | 19 |
+-------------------------------+--------------------+
| `matterport/Mask_RCNN`__ |TF| | 14 |
+-------------------------------+--------------------+
.. _maskrcnn-benchmark: https://github.com/facebookresearch/maskrcnn-benchmark/
.. _tensorpack: https://github.com/tensorpack/tensorpack/tree/master/examples/FasterRCNN
.. _mmdetection: https://github.com/open-mmlab/mmdetection/
.. _simpledet: https://github.com/TuSimple/simpledet/
.. _Detectron: https://github.com/facebookresearch/Detectron
__ https://github.com/matterport/Mask_RCNN/
.. |D2| image:: https://github.com/facebookresearch/detectron2/raw/master/.github/Detectron2-Logo-Horz.svg?sanitize=true
:height: 15pt
:target: https://github.com/facebookresearch/detectron2/
.. |PT| image:: https://pytorch.org/assets/images/logo-icon.svg
:width: 15pt
:height: 15pt
:target: https://pytorch.org
.. |TF| image:: https://static.nvidiagrid.net/ngc/containers/tensorflow.png
:width: 15pt
:height: 15pt
:target: https://tensorflow.org
.. |mxnet| image:: https://github.com/dmlc/web-data/raw/master/mxnet/image/mxnet_favicon.png
:width: 15pt
:height: 15pt
:target: https://mxnet.apache.org/
.. |C2| image:: https://caffe2.ai/static/logo.svg
:width: 15pt
:height: 15pt
:target: https://caffe2.ai
```
Details for each implementation:
* __Detectron2__: with release v0.1.2, run:
```
python tools/train_net.py --config-file configs/Detectron1-Comparisons/mask_rcnn_R_50_FPN_noaug_1x.yaml --num-gpus 8
```
* __mmdetection__: at commit `b0d845f`, run
```
./tools/dist_train.sh configs/mask_rcnn/mask_rcnn_r50_caffe_fpn_1x_coco.py 8
```
* __maskrcnn-benchmark__: use commit `0ce8f6f` with `sed -i ‘s/torch.uint8/torch.bool/g’ **/*.py; sed -i 's/AT_CHECK/TORCH_CHECK/g' **/*.cu`
to make it compatible with PyTorch 1.5. Then, run training with
```
python -m torch.distributed.launch --nproc_per_node=8 tools/train_net.py --config-file configs/e2e_mask_rcnn_R_50_FPN_1x.yaml
```
The speed we observed is faster than its model zoo, likely due to different software versions.
* __tensorpack__: at commit `caafda`, `export TF_CUDNN_USE_AUTOTUNE=0`, then run
```
mpirun -np 8 ./train.py --config DATA.BASEDIR=/data/coco TRAINER=horovod BACKBONE.STRIDE_1X1=True TRAIN.STEPS_PER_EPOCH=50 --load ImageNet-R50-AlignPadding.npz
```
* __SimpleDet__: at commit `9187a1`, run
```
python detection_train.py --config config/mask_r50v1_fpn_1x.py
```
* __Detectron__: run
```
python tools/train_net.py --cfg configs/12_2017_baselines/e2e_mask_rcnn_R-50-FPN_1x.yaml
```
Note that many of its ops run on CPUs, therefore the performance is limited.
* __matterport/Mask_RCNN__: at commit `3deaec`, apply the following diff, `export TF_CUDNN_USE_AUTOTUNE=0`, then run
```
python coco.py train --dataset=/data/coco/ --model=imagenet
```
Note that many small details in this implementation might be different
from Detectron's standards.
<details>
<summary>
(diff to make it use the same hyperparameters - click to expand)
</summary>
```diff
diff --git i/mrcnn/model.py w/mrcnn/model.py
index 62cb2b0..61d7779 100644
--- i/mrcnn/model.py
+++ w/mrcnn/model.py
@@ -2367,8 +2367,8 @@ class MaskRCNN():
epochs=epochs,
steps_per_epoch=self.config.STEPS_PER_EPOCH,
callbacks=callbacks,
- validation_data=val_generator,
- validation_steps=self.config.VALIDATION_STEPS,
+ #validation_data=val_generator,
+ #validation_steps=self.config.VALIDATION_STEPS,
max_queue_size=100,
workers=workers,
use_multiprocessing=True,
diff --git i/mrcnn/parallel_model.py w/mrcnn/parallel_model.py
index d2bf53b..060172a 100644
--- i/mrcnn/parallel_model.py
+++ w/mrcnn/parallel_model.py
@@ -32,6 +32,7 @@ class ParallelModel(KM.Model):
keras_model: The Keras model to parallelize
gpu_count: Number of GPUs. Must be > 1
"""
+ super().__init__()
self.inner_model = keras_model
self.gpu_count = gpu_count
merged_outputs = self.make_parallel()
diff --git i/samples/coco/coco.py w/samples/coco/coco.py
index 5d172b5..239ed75 100644
--- i/samples/coco/coco.py
+++ w/samples/coco/coco.py
@@ -81,7 +81,10 @@ class CocoConfig(Config):
IMAGES_PER_GPU = 2
# Uncomment to train on 8 GPUs (default is 1)
- # GPU_COUNT = 8
+ GPU_COUNT = 8
+ BACKBONE = "resnet50"
+ STEPS_PER_EPOCH = 50
+ TRAIN_ROIS_PER_IMAGE = 512
# Number of classes (including background)
NUM_CLASSES = 1 + 80 # COCO has 80 classes
@@ -496,29 +499,10 @@ if __name__ == '__main__':
# *** This training schedule is an example. Update to your needs ***
# Training - Stage 1
- print("Training network heads")
model.train(dataset_train, dataset_val,
learning_rate=config.LEARNING_RATE,
epochs=40,
- layers='heads',
- augmentation=augmentation)
-
- # Training - Stage 2
- # Finetune layers from ResNet stage 4 and up
- print("Fine tune Resnet stage 4 and up")
- model.train(dataset_train, dataset_val,
- learning_rate=config.LEARNING_RATE,
- epochs=120,
- layers='4+',
- augmentation=augmentation)
-
- # Training - Stage 3
- # Fine tune all layers
- print("Fine tune all layers")
- model.train(dataset_train, dataset_val,
- learning_rate=config.LEARNING_RATE / 10,
- epochs=160,
- layers='all',
+ layers='3+',
augmentation=augmentation)
elif args.command == "evaluate":
```
</details>

26
vton-api/preprocess/humanparsing/mhp_extension/detectron2/docs/notes/changelog.md Normal file
View File

@@ -0,0 +1,26 @@
# Change Log
### Releases
See release log at
[https://github.com/facebookresearch/detectron2/releases](https://github.com/facebookresearch/detectron2/releases).
### Notable Backward Incompatible Changes:
* 03/30/2020: Custom box head's `output_size` changed to `output_shape`.
* 02/14/2020,02/18/2020: Mask head and keypoint head now include logic for losses & inference. Custom heads
should overwrite the feature computation by `layers()` method.
* 11/11/2019: `detectron2.data.detection_utils.read_image` transposes images with exif information.
### Config Version Change Log
* v1: Rename `RPN_HEAD.NAME` to `RPN.HEAD_NAME`.
* v2: A batch of rename of many configurations before release.
### Silent Regression in Historical Versions:
We list a few silent regressions since they may silently produce incorrect results and will be hard to debug.
* 04/01/2020 - 05/11/2020: Bad accuracy if `TRAIN_ON_PRED_BOXES` is set to True.
* 03/30/2020 - 04/01/2020: ResNets are not correctly built.
* 12/19/2019 - 12/26/2019: Using aspect ratio grouping causes a drop in accuracy.
* release - 11/9/2019: Test time augmentation does not predict the last category.

83
vton-api/preprocess/humanparsing/mhp_extension/detectron2/docs/notes/compatibility.md Normal file
View File

@@ -0,0 +1,83 @@
# Compatibility with Other Libraries
## Compatibility with Detectron (and maskrcnn-benchmark)
Detectron2 addresses some legacy issues left in Detectron. As a result, their models
are not compatible:
running inference with the same model weights will produce different results in the two code bases.
The major differences regarding inference are:
- The height and width of a box with corners (x1, y1) and (x2, y2) is now computed more naturally as
width = x2 - x1 and height = y2 - y1;
In Detectron, a "+ 1" was added both height and width.
Note that the relevant ops in Caffe2 have [adopted this change of convention](https://github.com/pytorch/pytorch/pull/20550)
with an extra option.
So it is still possible to run inference with a Detectron2-trained model in Caffe2.
The change in height/width calculations most notably changes:
- encoding/decoding in bounding box regression.
- non-maximum suppression. The effect here is very negligible, though.
- RPN now uses simpler anchors with fewer quantization artifacts.
In Detectron, the anchors were quantized and
[do not have accurate areas](https://github.com/facebookresearch/Detectron/issues/227).
In Detectron2, the anchors are center-aligned to feature grid points and not quantized.
- Classification layers have a different ordering of class labels.
This involves any trainable parameter with shape (..., num_categories + 1, ...).
In Detectron2, integer labels [0, K-1] correspond to the K = num_categories object categories
and the label "K" corresponds to the special "background" category.
In Detectron, label "0" means background, and labels [1, K] correspond to the K categories.
- ROIAlign is implemented differently. The new implementation is [available in Caffe2](https://github.com/pytorch/pytorch/pull/23706).
1. All the ROIs are shifted by half a pixel compared to Detectron in order to create better image-feature-map alignment.
See `layers/roi_align.py` for details.
To enable the old behavior, use `ROIAlign(aligned=False)`, or `POOLER_TYPE=ROIAlign` instead of
`ROIAlignV2` (the default).
1. The ROIs are not required to have a minimum size of 1.
This will lead to tiny differences in the output, but should be negligible.
- Mask inference function is different.
In Detectron2, the "paste_mask" function is different and should be more accurate than in Detectron. This change
can improve mask AP on COCO by ~0.5% absolute.
There are some other differences in training as well, but they won't affect
model-level compatibility. The major ones are:
- We fixed a [bug](https://github.com/facebookresearch/Detectron/issues/459) in
Detectron, by making `RPN.POST_NMS_TOPK_TRAIN` per-image, rather than per-batch.
The fix may lead to a small accuracy drop for a few models (e.g. keypoint
detection) and will require some parameter tuning to match the Detectron results.
- For simplicity, we change the default loss in bounding box regression to L1 loss, instead of smooth L1 loss.
We have observed that this tends to slightly decrease box AP50 while improving box AP for higher
overlap thresholds (and leading to a slight overall improvement in box AP).
- We interpret the coordinates in COCO bounding box and segmentation annotations
as coordinates in range `[0, width]` or `[0, height]`. The coordinates in
COCO keypoint annotations are interpreted as pixel indices in range `[0, width - 1]` or `[0, height - 1]`.
Note that this affects how flip augmentation is implemented.
We will later share more details and rationale behind the above mentioned issues
about pixels, coordinates, and "+1"s.
## Compatibility with Caffe2
As mentioned above, despite the incompatibilities with Detectron, the relevant
ops have been implemented in Caffe2.
Therefore, models trained with detectron2 can be converted in Caffe2.
See [Deployment](../tutorials/deployment.md) for the tutorial.
## Compatibility with TensorFlow
Most ops are available in TensorFlow, although some tiny differences in
the implementation of resize / ROIAlign / padding need to be addressed.
A working conversion script is provided by [tensorpack FasterRCNN](https://github.com/tensorpack/tensorpack/tree/master/examples/FasterRCNN/convert_d2)
to run a standard detectron2 model in TensorFlow.

49
vton-api/preprocess/humanparsing/mhp_extension/detectron2/docs/notes/contributing.md Normal file
View File

@@ -0,0 +1,49 @@
# Contributing to detectron2
## Issues
We use GitHub issues to track public bugs and questions.
Please make sure to follow one of the
[issue templates](https://github.com/facebookresearch/detectron2/issues/new/choose)
when reporting any issues.
Facebook has a [bounty program](https://www.facebook.com/whitehat/) for the safe
disclosure of security bugs. In those cases, please go through the process
outlined on that page and do not file a public issue.
## Pull Requests
We actively welcome your pull requests.
However, if you're adding any significant features (e.g. > 50 lines), please
make sure to have a corresponding issue to discuss your motivation and proposals,
before sending a PR. We do not always accept new features, and we take the following
factors into consideration:
1. Whether the same feature can be achieved without modifying detectron2.
Detectron2 is designed so that you can implement many extensions from the outside, e.g.
those in [projects](https://github.com/facebookresearch/detectron2/tree/master/projects).
If some part is not as extensible, you can also bring up the issue to make it more extensible.
2. Whether the feature is potentially useful to a large audience, or only to a small portion of users.
3. Whether the proposed solution has a good design / interface.
4. Whether the proposed solution adds extra mental/practical overhead to users who don't
need such feature.
5. Whether the proposed solution breaks existing APIs.
When sending a PR, please do:
1. If a PR contains multiple orthogonal changes, split it to several PRs.
2. If you've added code that should be tested, add tests.
3. For PRs that need experiments (e.g. adding a new model or new methods),
you don't need to update model zoo, but do provide experiment results in the description of the PR.
4. If APIs are changed, update the documentation.
5. Make sure your code lints with `./dev/linter.sh`.
## Contributor License Agreement ("CLA")
In order to accept your pull request, we need you to submit a CLA. You only need
to do this once to work on any of Facebook's open source projects.
Complete your CLA here: <https://code.facebook.com/cla>
## License
By contributing to detectron2, you agree that your contributions will be licensed
under the LICENSE file in the root directory of this source tree.

10
vton-api/preprocess/humanparsing/mhp_extension/detectron2/docs/notes/index.rst Normal file
View File

@@ -0,0 +1,10 @@
Notes
======================================
.. toctree::
:maxdepth: 2
benchmarks
compatibility
contributing
changelog