Skip to content

Latest commit

 

History

History
executable file
·
191 lines (163 loc) · 7.2 KB

README_EN.md

File metadata and controls

executable file
·
191 lines (163 loc) · 7.2 KB
Raw

Simple Baselines for Human Pose Estimation and Tracking

Introduction

This is an paddle implementation of Simple Baselines for Human Pose Estimation and Tracking. This work provides baseline methods that are surprisingly simple and effective, thus helpful for inspiring and evaluating new ideas for the field. State-of-the-art results are achieved on challenging benchmarks.

Main Results

Results on MPII val

Arch Head Shoulder Elbow Wrist Hip Knee Ankle Mean Mean@0.1
256x256_pose_resnet_50_d256d256d256 96.658 95.482 89.041 83.587 88.333 84.625 80.090 88.808 34.241

Note:

  • Flip test is used.

Environment

The code is developed using python 3.7 on Ubuntu 16.04. NVIDIA GPUs are needed. The code is developed and tested using 1 NVIDIA V100 GPU cards.

Quick start

Installation

  1. Install paddle >= v2.1.2

  2. Clone this repo, and we'll call the directory that you cloned as ${POSE_ROOT}.

  3. Install dependencies:

    pip install -r requirements.txt

  4. Install coco dependencies:

    cd cocoapi/PythonAPI&&python3 setup.py install --user

  5. Init output(training model output directory) and log(visual log directory) directory:

    mkdir output 
mkdir log

    Your directory tree should look like this:

    ${POSE_ROOT}
├── data
├── experiments
├── images
├── lib
├── log
├── models
├── output
├── pose_estimation
├── README.md
└── requirements.txt

Data preparation

For MPII data, please download from MPII Human Pose Dataset. The original annotation files are in matlab format. We have converted them into json format, you also need to download them from OneDrive or GoogleDrive. Extract them under {POSE_ROOT}/data, and make them look like this:

${POSE_ROOT}
|-- data
`-- |-- mpii
    `-- |-- annot
        |   |-- gt_valid.mat
        |   |-- test.json
        |   |-- train.json
        |   |-- trainval.json
        |   `-- valid.json
        `-- images
            |-- 000001163.jpg
            |-- 000003072.jpg

Training on MPII

python pose_estimation/train.py \
    --cfg experiments/mpii/resnet50/256x256_d256x3_adam_lr1e-3.yaml

The training output info like:

2021-12-14 09:45:59     Epoch: [61][0/695]      Time 1.030s (1.030s)    Speed 31.1 samples/s    Data 0.819s (0.819s)    Lr 0.0010 (0.0010)      Loss 0.00045 (0.00045)  Accuracy 0.851 (0.851)
2021-12-14 09:46:49     Epoch: [61][50/695]     Time 1.268s (0.992s)    Speed 25.2 samples/s    Data 1.026s (0.800s)    Lr 0.0010 (0.0010)      Loss 0.00044 (0.00047)  Accuracy 0.878 (0.849)

Valid on MPII using pretrained models

pretained_model link: https://pan.baidu.com/s/1gaeyGPThltbyeoEQTx7RZQ valid code: sxfs

python pose_estimation/valid.py \
    --cfg experiments/mpii/resnet50/256x256_d256x3_adam_lr1e-3.yaml \
    --flip-test \
    --model-file output/mpii/pose_resnet_50/256x256_d256x3_adam_lr1e-3/model_best.pdparams

The valid info output format as follow:

Test: [0/93]	Time 5.032 (5.032)	Loss 0.0003 (0.0003)	Accuracy 0.946 (0.946)
Test: [50/93]	Time 0.254 (0.391)	Loss 0.0004 (0.0004)	Accuracy 0.901 (0.900)
| Arch | Head | Shoulder | Elbow | Wrist | Hip | Knee | Ankle | Mean | Mean@0.1 |
|---|---|---|---|---|---|---|---|---|---|
| 256x256_pose_resnet_50_d256d256d256 | 96.658 | 95.482 | 89.041 | 83.587 | 88.333 | 84.625 | 80.090 | 88.808 | 34.241 |

Infer on Test image using pretrained models

Note: You need to abtain object's location bbox firstly. details in infer.py line 138.

If there are multiple persons in images, detectors such as Faster R-CNN, SSD or others should be used first to crop them out. Because the simple baseline for human pose estimation is a top-down method.

python pose_estimation/infer.py --cfg experiments/mpii/resnet50/256x256_d256x3_adam_lr1e-3.yaml --img-file images/test.jpg  --model-file output/mpii/pose_resnet_50/256x256_d256x3_adam_lr1e-3/model_best.pdparams

Ultra-fast training-validation using mini datasets

The extracted mpii mini dataset containing 8 training images and 8 validation images is stored in the data folder in the repository. (The full dataset annotations and images are not included, just to verify the correctness of the training and validation for this project, please replace the entire dataset during subsequent training.) After installing the dependencies according to the above instructions, ultra-fast training and validation can be performed directly using the following commands:

  1. unzip mini datasets
unzip data/mpii.zip -d data/
  1. ultra-fast training-validation
python pose_estimation/train.py \
    --cfg experiments/mpii/resnet50/256x256_d256x3_adam_lr1e-3.yaml

output info on screen as follow:

=> load 15 samples
=> load 8 samples
2021-12-14 21:15:25     Epoch: [0][0/8] Time 1.835s (1.835s)    Speed 1.1 samples/s     Data 0.073s (0.073s)    Lr 0.0010 (0.0010)      Loss 0.55474 (0.55474)  Accuracy 0.000 (0.000)
Test: [0/1]     Time 1.839 (1.839)      Loss 0.0195 (0.0195)    Accuracy 0.023 (0.023)
| Arch | Head | Shoulder | Elbow | Wrist | Hip | Knee | Ankle | Mean | Mean@0.1 |
|---|---|---|---|---|---|---|---|---|---|
| 256x256_pose_resnet_50_d256d256d256 | 0.000 | 6.250 | 0.000 | 6.250 | 0.000 | 0.000 | 0.000 | 1.852 | 0.000 |
=> saving checkpoint to output/mpii/pose_resnet_50/256x256_d256x3_adam_lr1e-3

Export .ONNX using pretrained models

python pose_estimation/export.py \
    --cfg experiments/mpii/resnet50/256x256_d256x3_adam_lr1e-3.yaml \
    --model-file output/mpii/pose_resnet_50/256x256_d256x3_adam_lr1e-3/model_best.pdparams

The exporting info like:

2021-12-14 09:57:14 [INFO]      ONNX model saved in ./output/onnx/posenet.onnx
=> Model saved as: ./output/onnx/posenet.onnx
=> Done.

Deploy with exported ONNX models

Note: Only support for coco dataset now. The deploy.py is a example only.

python pose_estimation/deploy.py --img data/coco/images/xxx.jpg --model output/onnx/posenet.onnx --type ONNX --width 656 --height 384

Citation

Original project: https://github.com/microsoft/human-pose-estimation.pytorch

If you want use other dataset such as coco. Please check the orginal project's experiments setting.

The differences in .yaml file are as follows(You need to point out the depth of backbone and select lr_sche):

MODEL:
  NAME: pose_resnet
  BACKBONE: resnet
  DEPTH: 50

TRAIN:
  BATCH_SIZE: 128
  SHUFFLE: True
  BEGIN_EPOCH: 0
  END_EPOCH: 140
  RESUME: False
  OPTIMIZER: adam
  LR_TYPE: PolynomialDecay

If you want to cite the work in your research, please cite with:

@inproceedings{xiao2018simple,
    author={Xiao, Bin and Wu, Haiping and Wei, Yichen},
    title={Simple Baselines for Human Pose Estimation and Tracking},
    booktitle = {European Conference on Computer Vision (ECCV)},
    year = {2018}
}