etos-simplepose: A Python repository from etosworld

etos-simplepose

simple pose estimation.
for real time and high efficiency:
- only for single person pose estimation
- using lightweight CPM and Hourglass model
- Instead of normal convolution, inverted residuals (also known as Mobilenet V2) module has been used inside the model for real-time inference.
running this model with a normal pc webcam can achieve 15fps+ on ETOS system, with high accuracy.

Training

Dependencies:

Python3
TensorFlow >= 1.4

pip install -r requirements.txt

Dataset:

Training dataset available through google driver.

Unzip it will obtain the following file structure

$ tree -L 1 .
.
├── ai_challenger_train.json
├── ai_challenger_valid.json
├── train
└── valid

The traing dataset only contains single person images and it come from the competition of AI Challenger.

22446 training examples
1500 testing examples

the annotation was transferred into COCO format for using the data augument code from tf-pose-estimation respository. Beside, you also need to install cocoapi

Hyper-parameter

In training procedure, we use cfg file for passing the hyper-parameter.

Below is the content of mv2_hourglass.cfg.

[Train]
model: 'mv2_hourglass'
checkpoint: False
datapath: '~/data/ai_challenger'
imgpath: '~/data/'
visible_devices: '1'
multiprocessing_num: 16
max_epoch: 1000
lr: '0.001'
batchsize: 16
decay_rate: 0.95
input_width: 192
input_height: 192
n_kpoints: 14
scale: 4
identify_occlusion: False
modelpath: '~/data/trained/mv2_hourglass_deep/models'
logpath: '~/data/trained/mv2_hourglass_deep/log'
num_train_samples: 25000
per_update_tensorboard_step: 500
per_saved_model_step: 500
pred_image_on_tensorboard: False

The cfg not cover all the parameters of the model, there still have some parameters in the network_mv2_hourglass.py.

training

python3 train.py mv2_hourglass.cfg

Bechmark (PCKh)

Run the follow command to evaluate the value of your PCKh.

python3 src/benchmark.py --frozen_pb_path=hourglass/model-360000.pb \
--anno_json_path=~/data/ai_challenger/ai_challenger_valid.json \
--img_path=~/data \
--output_node_name=hourglass_out_3

pc webcam demo

python test.py

Reference

[1] Paper of Convolutional Pose Machines
[2] Paper of Stack Hourglass
[3] Paper of MobileNet V2
[4] Repository PoseEstimationForMobile
[5] Repository of tf-pose-estimation
[6] Devlope guide of TensorFlow Lite

etosworld/etos-simplepose