isarsoft/yolov4-triton-tensorrt
This repository deploys YOLOv4 as an optimized TensorRT engine to Triton Inference Server
5 Releases
Latest: 4y ago
Release v1.4.0v1.4.0Latest
philipp-schmidt·4y ago·January 12, 2022✨ What's New In v1.4.0
- Updated implementation for yolo layer with multiple changes and bugfixes
- Support for yolov4-tiny and yolov4-tiny-3l
- C++ client example (for previous versions <= v1.3.0, PRs welcome)
- Shared memory support for python client
- Support for TensorRT 8.0 (NGC 21.10)
- Performance improvements by ~19% (mainly due to new plugin and TRT 8)
- Support for Triton Inference Server 21.10
Release v1.3.0v1.3.0
philipp-schmidt·5y ago·September 25, 2020✨ What's New In v1.3.0
- Fixed confidence calculation (postprocessing)
- Update client to display confidence
- Update client to output amount of raw detection boxes
Release v1.2.0v1.2.0
philipp-schmidt·5y ago·September 22, 2020✨ What's New In v1.2.0
- Update to tensorrt container version 20.08
- Added detailed benchmarks
- Added python client with dummy mode, image mode and video mode
Release v1.1.0v1.1.0
philipp-schmidt·5y ago·August 17, 2020✨ What's New In v1.1.0
- Replaced MISH plugin layers by standard TensorRT activations resulting in 60% inference speedup
Release v1.0.0v1.0.0
philipp-schmidt·5y ago·August 16, 2020✨ Features
- Deployment to Triton Inference Server
- Execution with trtexec
- FP16 precision