10.1. Horizon Torch Samples¶
Horizon Torch Samples is a Pytorch-based deep-learning training tool provided by Horizon Robotics, which consists of Pytorch plugin and Horizon Torch Samples algorithm package.
Pytorch plugin is a set of quantization algorithm tools developed based on Pytorch, whose quantization algorithms are deeply coupled with Horizon processors, and the quantization models trained with this tool can be compiled and run normally on Horizon’s BPUs.
Horizon Torch Samples is an efficient and user-friendly algorithm tool based on the Pytorch and Pytorch plugin interfaces, which also provides state-of-the-art (SOTA) deep-learning models for common image tasks including classification, detection, segmentation, etc.
Horizon Torch Samples encapsulates and organizes all algorithm examples based on Horizon Algorithm Toolkit (HAT).
- 10.1.1. Overview
- 10.1.2. FRAMEWORK
- 10.1.3. TUTORIALS
- 10.1.3.1. Quantized Training
- 10.1.3.2. Writing Specifications of set_qconfig and Customization of qconfig
- 10.1.3.3. How to Turn on AMP
- 10.1.3.4. Model Compilation
- 10.1.3.5. Registration Mechanism
- 10.1.3.6. Config File
- 10.1.3.7. Config Construction of FCOS-EfficientNetB0
- 10.1.3.8. Overriding Config Parameters Using Commands
- 10.1.3.9. FLOPs Tool
- 10.1.3.10. Calibration
- 10.1.3.11. Start-up Method
- 10.1.4. EXAMPLES
- 10.1.4.1. Docker Image
- 10.1.4.2. Executing The Script
- 10.1.4.3. Multi-machine Instructions
- 10.1.4.4. MobileNetV1 Classification Model Training
- 10.1.4.5. RetinaNet Detection Model Training
- 10.1.4.6. YOLOv3 Detection Model Training
- 10.1.4.7. FCOS Detection Model Training
- 10.1.4.8. UNet Segmentation Model Training
- 10.1.4.9. PwcNet Optical Flow Prediction Model Training
- 10.1.4.10. PointPillars Detection Model Training
- 10.1.4.11. PointPillars Detection Model Training (No config)
- 10.1.4.12. PointPillars Detection Model Training (No config and HAT Trainer)
- 10.1.4.13. GaNet Lane Line Detection Model Training
- 10.1.4.14. FCOS3D Detection Model Training
- 10.1.4.15. Motr Multiple Object Track Model Training
- 10.1.4.16. CenterPoint Detection Model Training
- 10.1.4.17. StereoNet Binocular Depth Estimation Model Training
- 10.1.4.18. Bev Multi-task Model Training
- 10.1.4.19. Car Keypoint Detection Model Training
- 10.1.4.20. lidarMultiTask Model Training
- 10.1.4.21. DenseTNT Trajectory Prediction Model Training
- 10.1.5. Modelzoo
- 10.1.6. API REFERENCE