### Setup mmdeploy SDK for ONNX Inference

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/projects/rtmpose/README.md

Download and set up the mmdeploy SDK for ONNX inference on Linux. This involves downloading pre-compiled files, initializing the environment, and optionally installing OpenCV.

```shell
# Download pre-compiled files
wget https://github.com/open-mmlab/mmdeploy/releases/download/v1.0.0/mmdeploy-1.0.0-linux-x86_64-cxx11abi.tar.gz

# Unzip files
tar -xzvf mmdeploy-1.0.0-linux-x86_64-cxx11abi.tar.gz

# Go to the sdk folder
cd mmdeploy-1.0.0-linux-x86_64-cxx11abi

# Init environment
source set_env.sh

# If opencv 3+ is not installed on your system, execute the following command.
# If it is installed, skip this command
bash install_opencv.sh

# Compile executable programs
bash build_sdk.sh
```

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

### Install xtcocotools from Source (Alternative)

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/faq.md

If you encounter 'No matching distribution found for xtcocotools>=1.6', first install cython, then attempt installation from source. Ensure you have git installed.

```bash
pip install cython
```

```bash
git clone https://github.com/jin-s13/xtcocoapi
cd xtcocoapi
python setup.py install
```

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

### Install xtcocotools from Source

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/faq.md

If `pip install xtcocotools` fails, try cloning the repository and installing from source. Ensure you have git installed.

```bash
git clone https://github.com/jin-s13/xtcocoapi
cd xtcocoapi
python setup.py install
```

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

### Example: AMP Training Command

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/user_guides/train_and_test.md

A concrete example of a training command with AMP enabled, specifying a configuration file.

```shell
python tools/train.py configs/body_2d_keypoint/topdown_heatmap/coco/td-hm_res50_8xb64-210e_coco-256x192.py  --amp
```

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

### Install ONNXRuntime (CPU)

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/projects/rtmpose/examples/onnxruntime/README_CN.md

Install the CPU version of ONNXRuntime by downloading and extracting the binary. Set environment variables for the runtime directory and library path.

```bash
wget https://github.com/microsoft/onnxruntime/releases/download/v1.8.1/onnxruntime-linux-x64-1.8.1.tgz
tar -zxvf onnxruntime-linux-x64-1.8.1.tgz
export ONNXRUNTIME_DIR=$(pwd)/onnxruntime-linux-x64-1.8.1
export LD_LIBRARY_PATH=$ONNXRUNTIME_DIR/lib:$LD_LIBRARY_PATH
```

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

### Install TensorBoard

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/user_guides/train_and_test.md

Install the TensorBoard library using pip. This is a prerequisite for using TensorBoard visualization.

```shell
pip install tensorboard
```

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

### C++ SDK API Example

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/projects/rtmpose/README.md

A basic example demonstrating how to use the MMDeploy SDK's C++ API for pose detection.

```C++
#include "mmdeploy/detector.hpp"

#include "opencv2/imgcodecs/imgcodecs.hpp"
#include "utils/argparse.h"
#include "utils/visualize.h"

DEFINE_ARG_string(model, "Model path");
DEFINE_ARG_string(image, "Input image path");
DEFINE_STRING(device, "cpu", R"(Device name, e.g. "cpu", "cuda")");
DEFINE_STRING(output, "detector_output.jpg", "Output image path");

DEFINE_double(det_thr, .5, "Detection score threshold");

int main(int argc, char* argv[]) {
  if (!utils::ParseArguments(argc, argv)) {
    return -1;
  }

  cv::Mat img = cv::imread(ARGS_image);
  if (img.empty()) {
    fprintf(stderr, "failed to load image: %s\n", ARGS_image.c_str());
    return -1;
  }

  // construct a detector instance
  mmdeploy::Detector detector(mmdeploy::Model{ARGS_model}, mmdeploy::Device{FLAGS_device});

  // apply the detector, the result is an array-like class holding references to
  // `mmdeploy_detection_t`, will be released automatically on destruction
  mmdeploy::Detector::Result dets = detector.Apply(img);

  // visualize
  utils::Visualize v;
  auto sess = v.get_session(img);
  int count = 0;
  for (const mmdeploy_detection_t& det : dets) {
    if (det.score > FLAGS_det_thr) {  // filter bboxes
      sess.add_det(det.bbox, det.label_id, det.score, det.mask, count++);
    }
  }

  if (!FLAGS_output.empty()) {
    cv::imwrite(FLAGS_output, sess.get());
  }

  return 0;
}

```

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

### Install MMPose and Dependencies

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/projects/mmpose4aigc/README.md

Installs MMPose, MMDetection, and downloads necessary models. Ensure you have GCC and cmake installed.

```shell
pip install openmim
git clone https://github.com/open-mmlab/mmpose.git
cd mmpose
mim install -e .
mim install "mmdet>=3.0.0rc6"

bash download_models.sh
```

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

### Install mim tool

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/projects/rtmpose/examples/onnxruntime/README_CN.md

Install the 'mim' tool, which is used for downloading models.

```bash
pip install -U openmim
```

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

### Install MMPreTrain

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/configs/body_2d_keypoint/topdown_heatmap/coco/vitpose_coco.md

Install the required version of MMPreTrain using mim. This is a prerequisite for using ViTPose.

```shell
mim install 'mmpretrain>=1.0.0'
```

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

### Install TensorFlow

Source: https://github.com/idea-research/dwpose/blob/onnx/ControlNet-v1-1-nightly/annotator/zoe/zoedepth/models/base_models/midas_repo/mobile/ios/README.md

Install the TensorFlow library using pip. Ensure Python 3 is set as the default.

```shell
pip install tensorflow
```

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

### Example Inference with YOLOX-Pose

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/projects/yolox_pose/README.md

An example command to run inference on a specific image using a YOLOX-Pose configuration and checkpoint, saving visualization results.

```shell
python demo/inferencer_demo.py ../../tests/data/coco/000000000785.jpg \
    --pose2d configs/yolox-pose_s_8xb32-300e_coco.py \
    --pose2d-weights https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_s_8xb32-300e_coco-9f5e3924_20230321.pth \
    --scope mmyolo --vis-out-dir vis_results
```

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

### Install OpenVINO for OpenVINO Model

Source: https://github.com/idea-research/dwpose/blob/onnx/ControlNet-v1-1-nightly/annotator/zoe/zoedepth/models/base_models/midas_repo/README.md

Install the OpenVINO library using pip. This is required if you plan to use the OpenVINO model for inference on Intel CPUs.

```shell
pip install openvino
```

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

### Install MMPose Dependencies via MIM

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/demo/MMPose_Tutorial.ipynb

Installs MMEngine, MMCV, and MMDetection using the MIM package manager. This ensures compatibility and proper setup of core components.

```python
# install MMEngine, MMCV and MMDetection using MIM
%pip install -U openmim
!mim install mmengine
!mim install "mmcv>=2.0.0"
!mim install "mmdet>=3.0.0"
```

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

### General Configuration Example

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/user_guides/configs.md

This snippet shows the default general configurations including hooks, environment settings, visualizer, and logging.

```Python
# General
default_scope = 'mmpose'
default_hooks = dict(
    timer=dict(type='IterTimerHook'), # time the data processing and model inference
    logger=dict(type='LoggerHook', interval=50), # interval to print logs
    param_scheduler=dict(type='ParamSchedulerHook'), # update lr
    checkpoint=dict(
        type='CheckpointHook', interval=1, save_best='coco/AP', # interval to save ckpt
        rule='greater'), # rule to judge the metric
    sampler_seed=dict(type='DistSamplerSeedHook')) # set the distributed seed
env_cfg = dict(
    cudnn_benchmark=False, # cudnn benchmark flag
    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0), # num of opencv threads
    dist_cfg=dict(backend='nccl')) # distributed training backend
vis_backends = [dict(type='LocalVisBackend')] # visualizer backend
visualizer = dict( # Config of visualizer
    type='PoseLocalVisualizer',
    vis_backends=[dict(type='LocalVisBackend')],
    name='visualizer')
log_processor = dict( # Format, interval to log
    type='LogProcessor', window_size=50, by_epoch=True, num_digits=6)
log_level = 'INFO' # The level of logging
```

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

### Training Configuration Example

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/user_guides/configs.md

Sets parameters for resuming training, loading pre-trained weights, defining training epochs and validation intervals, and configuring learning rate schedules and optimizers.

```Python
resume = False # resume checkpoints from a given path, the training will be resumed from the epoch when the checkpoint's is saved
load_from = None # load models as a pre-trained model from a given path
train_cfg = dict(by_epoch=True, max_epochs=210, val_interval=10) # max epochs of training, interval to validate
param_scheduler = [
    dict( # warmup strategy
        type='LinearLR', begin=0, end=500, start_factor=0.001, by_epoch=False),
    dict( # scheduler
        type='MultiStepLR',
        begin=0,
        end=210,
        milestones=[170, 200],
        gamma=0.1,
        by_epoch=True)
]
optim_wrapper = dict(optimizer=dict(type='Adam', lr=0.0005)) # optimizer and initial lr
auto_scale_lr = dict(base_batch_size=512) # auto scale the lr according to batch size
```

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

### Setup mmdeploy SDK for TensorRT Inference

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/projects/rtmpose/README.md

Download and set up the mmdeploy SDK for TensorRT inference on Linux with CUDA support. This involves downloading pre-compiled files, initializing the environment, and optionally installing OpenCV.

```shell
# Download pre-compiled files
wget https://github.com/open-mmlab/mmdeploy/releases/download/v1.0.0/mmdeploy-1.0.0-linux-x86_64-cxx11abi-cuda11.3.tar.gz

# Unzip files
tar -xzvf mmdeploy-1.0.0-linux-x86_64-cxx11abi-cuda11.3.tar.gz

# Go to the sdk folder
cd mmdeploy-1.0.0-linux-x86_64-cxx11abi-cuda11.3

# Init environment
source set_env.sh

# If opencv 3+ is not installed on your system, execute the following command.
# If it is installed, skip this command
bash install_opencv.sh

# Compile executable programs
bash build_sdk.sh
```

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

### Launch TensorBoard Visualization

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/user_guides/train_and_test.md

Start the TensorBoard server to visualize training data. The log directory should point to the experiment's visualization data.

```shell
tensorboard --logdir ${WORK_DIR}/${TIMESTAMP}/vis_data
```

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

### Create a config file

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/demo/MMPose_Tutorial.ipynb

This section shows an example of a configuration file used in MMPose. It details the structure and parameters for setting up a pose estimation model.

```yaml
model =
  type='TopdownPoseEstimator'
  data_preprocessor=dict(type='PoseDataPreprocessor', mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375])
  backbone=dict(type='HRNet', in_channels=3, extra=dict(final_conv_kernel=1, with_norm_tree=True, num_branches=4, multiscale_mode='dfrac', module_name='modules.HRModule'), enc_modules=dict(modules=[dict(type='SpatialFusionModule', in_channels=32, num_modules=1, fuse_channels=32), dict(type='SpatialFusionModule', in_channels=64, num_modules=1, fuse_channels=64), dict(type='SpatialFusionModule', in_channels=128, num_modules=1, fuse_channels=128)], freeze_modules='all'), init_cfg=dict(type='Pretrained', checkpoint='https://download.openmmlab.com/mmpose/v1/models/hrnet/hrnet_w32-35412687_20220915.pth'))
  head=dict(type='TopdownHeatmapHead', num_classes=17, in_channels=32, loss=dict(type='KeypointLoss', name_value=0.5, use_target_weight=True), endpoint_decoder=dict(type='TopdownHeatmap', in_channels=32, out_channels=17), decoder=dict(type='TopdownHeatmapDecoder', num_classes=17, threshold=0.5, num_parts=17, topk=100), train_cfg=dict(num_epochs=210, lr_factor=0.1, warm_up_epochs=5), test_cfg=dict(flip_test=True, flip_mode='regress'))

dataset_type='CocoDataset'
log_level='INFO'
load_from='https://download.openmmlab.com/mmpose/v1/models/hrnet/hrnet_w32-35412687_20220915.pth'
resume=False

# dataset settings
backend_args=dict(backend='local')
train_dataloader=dict(batch_size=64, num_workers=10, persistent_workers=True, sampler=dict(type='DefaultSampler', shuffle=True), dataset=dict(type='CocoDataset', data_root='/home/PJLAB/jiangtao/Documents/git-clone/mmpose/data/coco/', data_prefix=dict(img='train2017/'), ann_file='annotations/person_keypoints_train2017.json', data_cfg=dict(type='bottomup', num_joints=17, dataset_class_name='coco', joint_pairs=[[1, 2], [2, 3], [3, 4], [1, 5], [5, 6], [6, 7], [1, 8], [8, 9], [9, 10], [1, 11], [11, 12], [12, 13], [1, 0], [0, 14], [14, 16], [0, 15], [15, 17]]), pipeline=[dict(type='LoadImage'), dict(type='LoadAnnotations', with_bbox=True, with_keypoints=True), dict(type='RandomHorizontalFlip', prob=0.5), dict(type='RandomFlip', direction='vertical', prob=0.5), dict(type='RandomScale', scale_factor=(-0.3, 0.3)), dict(type='RandomResolution', aspect_ratio_range=[0.75, 1.3333333333333333], scale_range=[0.5, 1.5]), dict(type='RandomCrop', crop_size=[256, 192], allow_negative_crop=True), dict(type='CropIndividualInstances', scale=3.0, allow_negative_crop=True), dict(type='PhotoMetricDistortion', contrast_prob=0.2, brightness_prob=0.2, saturation_prob=0.2, hue_prob=0.2), dict(type='GenerateKeypointTarget', sigma=3), dict(type='PackPoseInputs')]))

val_dataloader=dict(batch_size=1, num_workers=10, persistent_workers=True, sampler=dict(type='DefaultSampler', shuffle=False), dataset=dict(type='CocoDataset', data_root='/home/PJLAB/jiangtao/Documents/git-clone/mmpose/data/coco/', data_prefix=dict(img='val2017/'), ann_file='annotations/person_keypoints_val2017.json', data_cfg=dict(type='bottomup', num_joints=17, dataset_class_name='coco', joint_pairs=[[1, 2], [2, 3], [3, 4], [1, 5], [5, 6], [6, 7], [1, 8], [8, 9], [9, 10], [1, 11], [11, 12], [12, 13], [1, 0], [0, 14], [14, 16], [0, 15], [15, 17]]), pipeline=[dict(type='LoadImage'), dict(type='LoadAnnotations', with_bbox=True, with_keypoints=True), dict(type='PackPoseInputs')]))

test_dataloader=dict(batch_size=1, num_workers=10, persistent_workers=True, sampler=dict(type='DefaultSampler', shuffle=False), dataset=dict(type='CocoDataset', data_root='/home/PJLAB/jiangtao/Documents/git-clone/mmpose/data/coco/', data_prefix=dict(img='val2017/'), ann_file='annotations/person_keypoints_val2017.json', data_cfg=dict(type='bottomup', num_joints=17, dataset_class_name='coco', joint_pairs=[[1, 2], [2, 3], [3, 4], [1, 5], [5, 6], [6, 7], [1, 8], [8, 9], [9, 10], [1, 11], [11, 12], [12, 13], [1, 0], [0, 14], [14, 16], [0, 15], [15, 17]]), pipeline=[dict(type='LoadImage'), dict(type='LoadAnnotations', with_bbox=True, with_keypoints=True), dict(type='PackPoseInputs')]))

# runtime settings
vis_backends=[dict(type='LocalVisBackend')],
visualizer=dict(type='PoseLocalVisualizer', vis_backends=[dict(type='LocalVisBackend')], name='visualizer'),
log_processor=dict(type='LogProcessor', window_size=50, by_epoch=True),
log_handler=dict(type='FileHandler', init_kwargs=dict(filename='train.log')),

# training settings
train_cfg=dict(by_epoch=True, max_epochs=210, val_interval=10),
val_cfg=dict(),
test_cfg=dict())

```

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

### Verify MMPose Installation

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/installation.md

Check the installed MMPose version after installation.

```python
import mmpose
print(mmpose.__version__)
```

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

### Example: Run 2D Pose Estimation with MMDetection on Video

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/demo/docs/en/2d_wholebody_pose_demo.md

This command demonstrates running the 2D pose estimation demo with MMDetection on a video file. It uses a remote video URL as input and specifies output directory and display options.

```shell
python demo/topdown_demo_with_mmdet.py \
    demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py \
    https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
    configs/wholebody_2d_keypoint/topdown_heatmap/coco-wholebody/td-hm_hrnet-w48_dark-8xb32-210e_coco-wholebody-384x288.py \
    https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w48_coco_wholebody_384x288_dark-f5726563_20200918.pth \
    --input https://user-images.githubusercontent.com/87690686/137440639-fb08603d-9a35-474e-b65f-46b5c06b68d6.mp4 \
    --output-root vis_results/ --show
```

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

### Full Configuration Example with Codec

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/advanced_guides/codecs.md

A comprehensive configuration file snippet demonstrating the definition of a codec and its subsequent use in both the pipeline's GenerateTarget module and the model's head as a decoder.

```Python
# codec settings
codec = dict(type='RegressionLabel', input_size=(192, 256))                     ## definition ##

# model settings
model = dict(
    type='TopdownPoseEstimator',
    data_preprocessor=dict(
        type='PoseDataPreprocessor',
        mean=[123.675, 116.28, 103.53],
        std=[58.395, 57.12, 57.375],
        bgr_to_rgb=True),
    backbone=dict(
        type='ResNet',
        depth=50,
        init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50'),
    ),
    neck=dict(type='GlobalAveragePooling'),
    head=dict(
        type='RLEHead',
        in_channels=2048,
        num_joints=17,
        loss=dict(type='RLELoss', use_target_weight=True),
        decoder=codec),                                                         ## Head ##
    test_cfg=dict(
        flip_test=True,
        shift_coords=True,
    ))

# base dataset settings
dataset_type = 'CocoDataset'
data_mode = 'topdown'
data_root = 'data/coco/'

backend_args = dict(backend='local')

# pipelines
train_pipeline = [
    dict(type='LoadImage', backend_args=backend_args),
    dict(type='GetBBoxCenterScale'),
    dict(type='RandomFlip', direction='horizontal'),
    dict(type='RandomHalfBody'),
    dict(type='RandomBBoxTransform'),
    dict(type='TopdownAffine', input_size=codec['input_size']),
    dict(type='GenerateTarget', encoder=codec),   ## Generate Target ##
    dict(type='PackPoseInputs')
]
test_pipeline = [
    dict(type='LoadImage', backend_args=backend_args),
    dict(type='GetBBoxCenterScale'),
    dict(type='TopdownAffine', input_size=codec['input_size']),
    dict(type='PackPoseInputs')
]
```

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

### Install Tree Package

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/demo/MMPose_Tutorial.ipynb

This command installs the 'tree' package, which is useful for visualizing directory structures. It shows the output of the installation process.

```bash
Reading package lists... 
Building dependency tree... 
Reading state information... 
The following package was automatically installed and is no longer required: 
  libnvidia-common-460
Use 'apt autoremove' to remove it.
The following NEW packages will be installed: 
  tree
0 upgraded, 1 newly installed, 0 to remove and 32 not upgraded.
Need to get 40.7 kB of archives.
After this operation, 105 kB of additional disk space will be used.
Get:1 http://archive.ubuntu.com/ubuntu bionic/universe amd64 tree amd64 1.7.0-5 [40.7 kB]
Fetched 40.7 kB in 0s (161 kB/s)
Selecting previously unselected package tree.
(Reading database ... 155685 files and directories currently installed.)
Preparing to unpack .../tree_1.7.0-5_amd64.deb ... 
Unpacking tree (1.7.0-5) ...
Setting up tree (1.7.0-5) ...
Processing triggers for man-db (2.8.3-2ubuntu0.1) ...
data/coco_tiny
├── images
│   ├── 000000012754.jpg
│   ├── 000000017741.jpg
│   ├── 000000019157.jpg
│   ├── 000000019523.jpg
│   ├── 000000019608.jpg
│   ├── 000000022816.jpg
│   ├── 000000031092.jpg
│   ├── 000000032124.jpg
│   ├── 000000037209.jpg
│   ├── 000000050713.jpg
│   ├── 000000057703.jpg
│   ├── 000000064909.jpg
│   ├── 000000076942.jpg
│   ├── 000000079754.jpg
│   ├── 000000083935.jpg
│   ├── 000000085316.jpg
│   ├── 000000101013.jpg
│   ├── 000000101172.jpg
│   ├── 000000103134.jpg
│   ├── 000000103163.jpg
│   ├── 000000105647.jpg
│   ├── 000000107960.jpg
│   ├── 000000117891.jpg
│   ├── 000000118181.jpg
│   ├── 000000120021.jpg
│   ├── 000000128119.jpg
│   ├── 000000143908.jpg
│   ├── 000000145025.jpg
│   ├── 000000147386.jpg
│   ├── 000000147979.jpg
│   ├── 000000154222.jpg
│   ├── 000000160190.jpg
│   ├── 000000161112.jpg
│   ├── 000000175737.jpg
│   ├── 000000177069.jpg
│   ├── 000000184659.jpg
│   ├── 000000209468.jpg
│   ├── 000000210060.jpg
│   ├── 000000215867.jpg
│   ├── 000000216861.jpg
│   ├── 000000227224.jpg
│   ├── 000000246265.jpg
│   ├── 000000254919.jpg
│   ├── 000000263687.jpg
│   ├── 000000264628.jpg
│   ├── 000000268927.jpg
│   ├── 000000271177.jpg
│   ├── 000000275219.jpg
│   ├── 000000277542.jpg
│   ├── 000000279140.jpg
│   ├── 000000286813.jpg
│   ├── 000000297980.jpg
│   ├── 000000301641.jpg
│   ├── 000000312341.jpg
│   ├── 000000325768.jpg
│   ├── 000000332221.jpg
│   ├── 000000345071.jpg
│   ├── 000000346965.jpg
│   ├── 000000347836.jpg
│   ├── 000000349437.jpg
│   ├── 000000360735.jpg
│   ├── 000000362343.jpg
│   ├── 000000364079.jpg
│   ├── 000000364113.jpg
│   ├── 000000386279.jpg
│   ├── 000000386968.jpg
│   ├── 000000388619.jpg
│   ├── 000000390137.jpg
│   ├── 000000390241.jpg
│   ├── 000000390298.jpg
│   ├── 000000390348.jpg
│   ├── 000000398606.jpg
│   ├── 000000400456.jpg
│   ├── 000000402514.jpg
│   ├── 000000403255.jpg
│   ├── 000000403432.jpg
│   ├── 000000410350.jpg
│   ├── 000000453065.jpg
│   ├── 000000457254.jpg
│   ├── 000000464153.jpg
│   ├── 000000464515.jpg
│   ├── 000000465418.jpg
│   ├── 000000480591.jpg
│   ├── 000000484279.jpg
│   ├── 000000494014.jpg
│   ├── 000000515289.jpg
│   ├── 000000516805.jpg
│   ├── 000000521994.jpg
│   ├── 000000528962.jpg
│   ├── 000000534736.jpg
│   ├── 000000535588.jpg
│   ├── 000000537548.jpg
│   ├── 000000553698.jpg
│   ├── 000000555622.jpg
│   ├── 000000566456.jpg
│   ├── 000000567171.jpg
│   └── 000000568961.jpg
train.json
└── val.json

1 directory, 99 files

```

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

### CheckpointHook Configuration Example

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/user_guides/configs.md

Demonstrates how to configure CheckpointHook with specific save_best metric, rule, and max_keep_ckpts.

```Python
default_hooks = dict(checkpoint=dict(save_best='PCK', rule='greater', max_keep_ckpts=1))
```

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

### Install MMPose as a Python Package

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/installation.md

Install MMPose using pip for use as a dependency or third-party package. Ensure you have mim installed.

```shell
mim install "mmpose>=1.1.0"
```

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

### Data Configuration Example

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/user_guides/configs.md

Defines data storage backend, dataset type, data root, codec for target generation, and training/testing pipelines. Includes dataloader configurations for training and validation.

```Python
backend_args = dict(backend='local') # data storage backend
dataset_type = 'CocoDataset' # name of dataset
data_mode = 'topdown' # type of the model
data_root = 'data/coco/' # root of the dataset
 # config of codec，to generate targets and decode preds into coordinates
codec = dict(
    type='MSRAHeatmap',
    input_size=(192, 256),
    heatmap_size=(48, 64),
    sigma=2)
train_pipeline = [ # data aug in training
    dict(type='LoadImage', backend_args=backend_args), # image loading
    dict(type='GetBBoxCenterScale'), # calculate center and scale of bbox
    dict(type='RandomBBoxTransform'), # config of scaling, rotation and shifing
    dict(type='RandomFlip', direction='horizontal'), # config of random flipping
    dict(type='RandomHalfBody'), # config of half-body aug
    dict(type='TopdownAffine', input_size=codec['input_size']), # update inputs via transform matrix
    dict(
        type='GenerateTarget', # generate targets via transformed inputs
        # typeof targets
        encoder=codec, # get encoder from codec
    dict(type='PackPoseInputs') # pack targets
]
test_pipeline = [ # data aug in testing
    dict(type='LoadImage', backend_args=backend_args), # image loading
    dict(type='GetBBoxCenterScale'), # calculate center and scale of bbox
    dict(type='TopdownAffine', input_size=codec['input_size']), # update inputs via transform matrix
    dict(type='PackPoseInputs') # pack targets
]
train_dataloader = dict(
    batch_size=64, # batch size of each single GPU during training
    num_workers=2, # workers to pre-fetch data for each single GPU
    persistent_workers=True, # workers will stay around (with their state) waiting for another call into that dataloader.
    sampler=dict(type='DefaultSampler', shuffle=True), # data sampler, shuffle in traning
    dataset=dict(
        type=dataset_type , # name of dataset
        data_root=data_root, # root of dataset
        data_mode=data_mode, # type of the model
        ann_file='annotations/person_keypoints_train2017.json', # path to annotation file
        data_prefix=dict(img='train2017/'), # path to images
        pipeline=train_pipeline
    ))
val_dataloader = dict(
    batch_size=32, # batch size of each single GPU during validation
    num_workers=2, # workers to pre-fetch data for each single GPU
    persistent_workers=True, # workers will stay around (with their state) waiting for another call into that dataloader.
    drop_last=False,
    sampler=dict(type='DefaultSampler', shuffle=False), # data sampler
    dataset=dict(
        type=dataset_type , # name of dataset
        data_root=data_root, # root of dataset
        data_mode=data_mode, # type of the model
        ann_file='annotations/person_keypoints_val2017.json', # path to annotation file
        bbox_file=
        'data/coco/person_detection_results/COCO_val2017_detections_AP_H_56_person.json', # bbox file use for evaluation
        data_prefix=dict(img='val2017/'), # path to images
        test_mode=True,
        pipeline=test_pipeline
    ))
test_dataloader = val_dataloader # use val as test by default
```

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

### Install MMPose Dependencies with MIM

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/installation.md

Install MMEngine, MMCV, and optionally MMDetection using the MIM package manager. Ensure MIM is installed or updated first.

```shell
pip install -U openmim
```

```shell
mim install mmengine
```

```shell
mim install "mmcv>=2.0.1"
```

```shell
mim install "mmdet>=3.1.0"
```

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

### Install PyTorch on GPU

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/installation.md

Install PyTorch and torchvision using conda. This command automatically installs the latest compatible PyTorch and cudatoolkit versions. Verify compatibility with your environment.

```shell
conda install pytorch torchvision -c pytorch
```

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

### Example: Run 2D Pose Estimation with MMDetection on Image

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/demo/docs/en/2d_wholebody_pose_demo.md

An example command for running the 2D pose estimation demo with MMDetection on a specific image. It includes paths to MMDetection and MMPose configurations and checkpoints, input image, and output root.

```shell
python demo/topdown_demo_with_mmdet.py \
    demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py \
    https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
    configs/wholebody_2d_keypoint/topdown_heatmap/coco-wholebody/td-hm_hrnet-w48_dark-8xb32-210e_coco-wholebody-384x288.py \
    https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w48_coco_wholebody_384x288_dark-f5726563_20200918.pth \
    --input tests/data/coco/000000196141.jpg \
    --output-root vis_results/ --show
```

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

### Clone MMPose and Install Dependencies

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/demo/MMPose_Tutorial.ipynb

This snippet clones the MMPose repository, navigates into the directory, and installs all required packages listed in requirements.txt. It then installs MMPose in editable mode for local development.

```python
!git clone https://github.com/open-mmlab/mmpose.git
# The master branch is version 1.x 
%cd mmpose
%pip install -r requirements.txt
%pip install -v -e .
# "-v" means verbose, or more output
# "-e" means installing a project in editable mode,
# thus any local modifications made to the code will take effect without reinstallation.
```

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

### Install Requirements

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/projects/rtmpose/examples/onnxruntime/README_CN.md

Install the necessary Python dependencies for running the inference script.

```bash
pip install -r requirements.txt
```

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

### Download Config and Checkpoint Files

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/installation.md

Download necessary configuration and checkpoint files for running inference demos. Specify the model config and destination folder.

```shell
mim download mmpose --config td-hm_hrnet-w48_8xb32-210e_coco-256x192  --dest .
```

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

### Print Configuration

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/user_guides/configs.md

Use this command to inspect the complete configuration of a given config file.

```Bash
python tools/analysis/print_config.py /PATH/TO/CONFIG
```

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

### Install PyTorch on CPU

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/installation.md

Install PyTorch and torchvision for CPU-only platforms using conda.

```shell
conda install pytorch torchvision cpuonly -c pytorch
```

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

### Example: 2D Human Pose Top-Down Demo with MMDetection

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/demos.md

Example execution of the top-down demo using MMDetection for detection and MMpose for pose estimation on an image. Includes visualization.

```shell
python demo/topdown_demo_with_mmdet.py \
    demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py \
    https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
    configs/body_2d_keypoint/topdown_heatmap/coco/td-hm_hrnet-w32_8xb64-210e_coco-256x192.py \
    https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w32_coco_256x192-c78dce93_20200708.pth \
    --input tests/data/coco/000000197388.jpg --show --draw-heatmap \
    --output-root vis_results/
```

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

### Install MMEngine with Pip

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/docs/en/installation.md

Use this command to install MMEngine using pip when not using MIM.

```shell
pip install mmengine
```

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

### Run 2D Face Image Demo with Specific Models

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/demo/docs/en/2d_face_demo.md

Run the demo with specific pre-trained models for face detection (YOLOX-S) and face keypoint estimation (HRNetV2-W18). This example visualizes keypoints and heatmaps for a given image.

```shell
python demo/topdown_demo_with_mmdet.py \
    demo/mmdetection_cfg/yolox-s_8xb8-300e_coco-face.py \
    https://download.openmmlab.com/mmpose/mmdet_pretrained/yolo-x_8xb8-300e_coco-face_13274d7c.pth \
    configs/face_2d_keypoint/topdown_heatmap/aflw/td-hm_hrnetv2-w18_8xb64-60e_aflw-256x256.py \
    https://download.openmmlab.com/mmpose/face/hrnetv2/hrnetv2_w18_aflw_256x256-f2bbc62b_20210125.pth \
    --input tests/data/cofw/001766.jpg \
    --show --draw-heatmap
```

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

### C++ SDK API Example

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/projects/rtmpose/README.md

This C++ code snippet shows how to initialize and use the MMDeploy Detector for inference. It includes argument parsing for model and image paths, device selection, and visualization of detection results.

```cpp
#include "mmdeploy/detector.hpp"

#include "opencv2/imgcodecs/imgcodecs.hpp"
#include "utils/argparse.h"
#include "utils/visualize.h"

DEFINE_ARG_string(model, "Model path");
DEFINE_ARG_string(image, "Input image path");
DEFINE_string(device, "cpu", R"(Device name, e.g. "cpu", "cuda")");
DEFINE_string(output, "detector_output.jpg", "Output image path");

DEFINE_double(det_thr, .5, "Detection score threshold");

int main(int argc, char* argv[]) {
  if (!utils::ParseArguments(argc, argv)) {
    return -1;
  }

  cv::Mat img = cv::imread(ARGS_image);
  if (img.empty()) {
    fprintf(stderr, "failed to load image: %s\n", ARGS_image.c_str());
    return -1;
  }

  // construct a detector instance
  mmdeploy::Detector detector(mmdeploy::Model{ARGS_model}, mmdeploy::Device{FLAGS_device});

  // apply the detector, the result is an array-like class holding references to
  // `mmdeploy_detection_t`, will be released automatically on destruction
  mmdeploy::Detector::Result dets = detector.Apply(img);

  // visualize
  utils::Visualize v;
  auto sess = v.get_session(img);
  int count = 0;
  for (const mmdeploy_detection_t& det : dets) {
    if (det.score > FLAGS_det_thr) {  // filter bboxes
      sess.add_det(det.bbox, det.label_id, det.score, det.mask, count++);
    }
  }

  if (!FLAGS_output.empty()) {
    cv::imwrite(FLAGS_output, sess.get());
  }

  return 0;
}

```

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

### Run 2D Animal Pose Demo on CPU

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/demo/docs/en/2d_animal_demo.md

This command demonstrates how to run the 2D animal pose estimation demo using the CPU instead of a GPU. Append the `--device cpu` argument to the standard command.

```shell
python demo/topdown_demo_with_mmdet.py \
    demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py \
    https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
    configs/animal_2d_keypoint/topdown_heatmap/animalpose/td-hm_hrnet-w32_8xb64-210e_animalpose-256x256.py \
    https://download.openmmlab.com/mmpose/animal/hrnet/hrnet_w32_animalpose_256x256-1aa7f075_20210426.pth \
    --input tests/data/animalpose/ca110.jpeg \
    --show --draw-heatmap --det-cat-id=15 --device cpu
```

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

### Install Python Scripts

Source: https://github.com/idea-research/dwpose/blob/onnx/ControlNet-v1-1-nightly/annotator/zoe/zoedepth/models/base_models/midas_repo/ros/midas_cpp/CMakeLists.txt

Installs Python scripts to the specified destination within the ROS package.

```cmake
catkin_install_python(PROGRAMS
  scripts/my_python_script
  DESTINATION ${CATKIN_PACKAGE_BIN_DESTINATION}
)
```

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

### Install MMPose Dependencies

Source: https://github.com/idea-research/dwpose/blob/onnx/INSTALL.md

Installs PyTorch with specific CUDA version and other requirements for the MMPose environment.

```bash
pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html
pip install -r requirements.txt
```

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

### Run 2D Face Demo on CPU

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/demo/docs/en/2d_face_demo.md

Execute the demo script using the CPU for processing. Set the `--device` argument to 'cpu'. This is useful when a GPU is not available.

```shell
python demo/topdown_demo_with_mmdet.py \
    demo/mmdetection_cfg/yolox-s_8xb8-300e_coco-face.py \
    https://download.openmmlab.com/mmpose/mmdet_pretrained/yolo-x_8xb8-300e_coco-face_13274d7c.pth \
    configs/face_2d_keypoint/topdown_heatmap/aflw/td-hm_hrnetv2-w18_8xb64-60e_aflw-256x256.py \
    https://download.openmmlab.com/mmpose/face/hrnetv2/hrnetv2_w18_aflw_256x256-f2bbc62b_20210125.pth \
    --input tests/data/cofw/001766.jpg \
    --show --draw-heatmap --device=cpu
```

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

### MMPose Configuration Example

Source: https://github.com/idea-research/dwpose/blob/onnx/mmpose/demo/MMPose_Tutorial.ipynb

This snippet shows a configuration for a pose estimation model, including dataset and pipeline settings. It is used to define the model architecture, data loading, and evaluation metrics.

```python
input_size=(192, 256),
                    heatmap_size=(48, 64),
                    sigma=2)),
            dict(type='PackPoseInputs')
        ]))
val_dataloader = dict(
    batch_size=16,
    num_workers=2,
    persistent_workers=True,
    drop_last=False,
    sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
    dataset=dict(
        type='TinyCocoDataset',
        data_root='data/coco_tiny',
        data_mode='topdown',
        ann_file='val.json',
        bbox_file=None,
        data_prefix=dict(img='images/'),
        test_mode=True,
        pipeline=[
            dict(type='LoadImage', file_client_args=dict(backend='disk')),
            dict(type='GetBBoxCenterScale'),
            dict(type='TopdownAffine', input_size=(192, 256)),
            dict(type='PackPoseInputs')
        ]))
test_dataloader = dict(
    batch_size=16,
    num_workers=2,
    persistent_workers=True,
    drop_last=False,
    sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
    dataset=dict(
        type='TinyCocoDataset',
        data_root='data/coco_tiny',
        data_mode='topdown',
        ann_file='val.json',
        bbox_file=None,
        data_prefix=dict(img='images/'),
        test_mode=True,
        pipeline=[
            dict(type='LoadImage', file_client_args=dict(backend='disk')),
            dict(type='GetBBoxCenterScale'),
            dict(type='TopdownAffine', input_size=(192, 256)),
            dict(type='PackPoseInputs')
        ]))
val_evaluator = dict(type='PCKAccuracy')
test_evaluator = dict(type='PCKAccuracy')
work_dir = 'work_dirs/hrnet_w32_coco_tiny_256x192'
randomness = dict(seed=0)
```

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

### Install Other Files

Source: https://github.com/idea-research/dwpose/blob/onnx/ControlNet-v1-1-nightly/annotator/zoe/zoedepth/models/base_models/midas_repo/ros/midas_cpp/CMakeLists.txt

Installs miscellaneous files, such as launch or bag files, to the package's share destination.

```cmake
install(FILES
  # myfile1
  # myfile2
  DESTINATION ${CATKIN_PACKAGE_SHARE_DESTINATION}
)
```

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

### Install Executable Target

Source: https://github.com/idea-research/dwpose/blob/onnx/ControlNet-v1-1-nightly/annotator/zoe/zoedepth/models/base_models/midas_repo/ros/midas_cpp/CMakeLists.txt

Installs a specific executable target to the ROS package's binary destination.

```cmake
install(TARGETS ${PROJECT_NAME}_node
  RUNTIME DESTINATION ${CATKIN_PACKAGE_BIN_DESTINATION}
)
```