The project will implement human keypoint detection algorithms, including YOLOv5 model implementation human detection (Person Detection), HRNet, LiteHRNet and Mobilenet-v2 model implementation human keypoint detection.
In order to facilitate subsequent model engineering and Android platform deployment, the project supports high-accuracy HRNet detection model, lightweight model LiteHRNet and Mobilenet model training and testing, and provides Python/C++ programming language /Android multiple versions;
This project mainly uses COCO data set and MPII data set
Inference code demo.py supports picture, video and camera testing
We have Python, C++ , Andriod version code
If you have any questions or need the full code or data or need customized development requirements, you can contact me Email Mail me : slowlon@foxmail.com
Using Top-Down (top-down) method, using YOLOv5 model to achieve human object detection, using HRNet for human key point detection (human pose estimation)
The project supports HRNet and lightweight model LiteHRNet and Mobilenet model training, and provides the corresponding configuration file; you need to modify the data path of the corresponding configuration file; this article takes training HRNet-w32 as an example, its configuration file is in configs/coco/hrnet/w32_adam_192_192 .yaml, modify the file training data set path TRAIN_FILE (support multiple data set training) and test data set TEST_FILE data path for your local data path, other parameters can be kept default, as follows:
After modifying the configuration file, you can start preparing for training:
Train high accuracy models HRNet-w48 or HRNet-w32
The following table gives HRNet, as well as lightweight models LiteHRNet and Mobilenet calculation amount and parameter number, and its detection accuracy AP; high accuracy detection model HRNet-w32, AP can reach 0.7585, but its parameters and calculation amount is relatively large, not suitable for mobile end deployment; LiteHRNet18 and Mobilenet-v2 parameters and calculation amount is relatively small, suitable for mobile end deployment; Although LiteHRNet18 theoretical calculation amount and parameter amount is lower than Mobilenet-v2, but in the actual test, it is found that Mobilenet-v2 runs faster. Lightweight Mobile-v2 model can achieve real-time detection effect on ordinary Android mobile phones, CPU (4 threads) about 50ms, GPU about 30ms, basically meet the performance requirements of business
Android version
Inference code demo.py supports picture, video and camera testing
We have Python, C++ , Andriod version code
If you have any questions or need the full code or data or need customized development requirements, you can contact me Email Mail me : slowlon@foxmail.com