https://github.com/TrojanXu/yolov5-tensorrt

A tensorrt implementation of yolov5: https://github.com/ultralytics/yolov5
https://github.com/TrojanXu/yolov5-tensorrt

Last synced: about 1 month ago
JSON representation

A tensorrt implementation of yolov5: https://github.com/ultralytics/yolov5

• Host: GitHub
• URL: https://github.com/TrojanXu/yolov5-tensorrt
• Owner: TrojanXu
• License: apache-2.0
• Created: 2020-06-17T01:02:45.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2020-10-04T07:51:07.000Z (about 4 years ago)
• Last Synced: 2024-08-02T01:19:48.400Z (5 months ago)
• Language: Python
• Size: 13.1 MB
• Stars: 191
• Watchers: 9
• Forks: 46
• Open Issues: 22
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-yolo-object-detection - TrojanXu/yolov5-tensorrt - tensorrt?style=social"/> : A tensorrt implementation of yolov5. (Lighter and Deployment Frameworks)
• awesome-cuda-triton-hpc - TrojanXu/yolov5-tensorrt - tensorrt?style=social"/> : A tensorrt implementation of yolov5. (Applications)
• awesome-cuda-triton-hpc - TrojanXu/yolov5-tensorrt - tensorrt?style=social"/> : A tensorrt implementation of yolov5. (Applications)

README

        
# yolov5-tensorrt

A tensorrt implementation of yolov5: https://github.com/ultralytics/yolov5

# requirement

Please use torch>=1.6.0 + onnx>=1.6.0 + TRT 7.1+ (fix upsample issue) to run the sample code  

onnx-simplifier-0.2.16

# The code

Add newly implemented upsample to get this working with current combination of onnx and tensorrt.  

0. prepare above mentioned environment.

1. git clone && git submodule update --init

2. download weights file (use yolov5/models/export.py)

3. python main.py to run the benchmark

4. Generally, for image of size 640*640, using batchsize=1, the speedup is 4x on V100.

# Updates

- 20201004 update to track yolov5 - v3.0 release. download model file from official websites please.

# TODO

- [x] NMS support

- [ ] dynamic shape or dynamic batchsize support (**won't implement soon because onnx-simplifier only supports fixed shape**)

- [ ] FP16 numerical issue and performance investigation

- [ ] Benchmark