LiteHRNet:Pose Estimation
LiteHRNet is a lightweight high-resolution network designed for real-time computer vision tasks like pose estimation. Optimized from the HRNet architecture, it reduces computational redundancy and parameters while preserving multi-scale feature fusion, enabling efficient deployment on mobile or resource-constrained devices. Its core strength lies in high-resolution detail retention and efficient feature extraction, supporting applications such as human pose estimation, facial landmark detection, and action recognition in scenarios like smart surveillance, mobile fitness coaching, and AR interaction. Challenges include balancing accuracy with speed, maintaining multi-scale information in compact models, and ensuring cross-device compatibility.
Source model
- Input shape: 1x3x256x192
- Number of parameters: 1.08M
- Model size: 13.82M
- Output shape: 17x2, 17, 17x64x48
The source model can be found here
Performance Reference
Please search model by model name in Model Farm
Inference & Model Conversion
Please search model by model name in Model Farm
License
Source Model: APACHE-2.0
Deployable Model: APLUX-MODEL-FARM-LICENSE