https://github.com/developer0hye/yolov8-tensorrt-inference-docker-image
Simply run your YOLOv8 faster by using TensorRT on a docker container
https://github.com/developer0hye/yolov8-tensorrt-inference-docker-image
Last synced: 6 months ago
JSON representation
Simply run your YOLOv8 faster by using TensorRT on a docker container
- Host: GitHub
- URL: https://github.com/developer0hye/yolov8-tensorrt-inference-docker-image
- Owner: developer0hye
- License: mit
- Created: 2024-01-28T12:32:54.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2024-01-30T15:30:40.000Z (about 2 years ago)
- Last Synced: 2025-09-27T09:59:24.610Z (6 months ago)
- Language: Python
- Homepage:
- Size: 9.92 MB
- Stars: 4
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# yolov8-tensorrt-python-docker-image
run your yolov8 faster simply using tensorrt on docker image
# Setup
## 1. Build an image and run a container
```bash
docker build . -t yolov8trt:1.0
docker run --name yolov8trt-container -it --runtime=nvidia --gpus all --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 yolov8trt:1.0
```
If you want to access your dataset on a container, mount a volume using `-v` flag.
## 2. Upload your model and data on a container
```bash
docker cp {your_model.pt} yolov8trt-container:/app/
docker cp {your_data} yolov8trt-container:/app/
```
## 3. Export your model with TensorRT
```bash
yolo export model={your model}.pt format=engine device=0 half=True dynamic=True
```
Enter `yolov8x` in {your model} for simple demo
## 4. Inference with TensorRT engine
```bash
python app.py --source {dataset path} --model {model path}
```
Enter `sample.mp4` in {dataset path} for simple demo
Enter `yolov8x.engine` in {model path} for simple demo
### Output Example
{the name of input video or image}.txt
```
frame_idx(start from 1), class index, confidence score, top left x, top left y, bottom right x, bottom right y
...
```
## Experiment
### Hardware
| Component | Specific Model |
| ------------- | ------------- |
| CPU | Intel i5-10400 |
| GPU | NVIDIA RTX 3070 |
| RAM | 32GB |
### Total Processing Time
- Input: sample.mp4
- Resoultion: 1080 x 1920
- Frame Rate: 30 FPS
- Total Frames: 682
- Duration: 22.73 sec
| Model | Inference Engine | Total Processing Time (sec) |
| ------------- | ------------- | ------------- |
| YOLOv8x | PyTorch CUDA | 26.46 |
| YOLOv8x | TensorRT | 20.35 |
| YOLOv8l | PyTorch CUDA | 26.42 |
| YOLOv8l | TensorRT | 19.17 |
| YOLOv8m | PyTorch CUDA | 24.88 |
| YOLOv8m | TensorRT |20.18 |
| YOLOv8s | PyTorch CUDA | 23.53 |
| YOLOv8s | TensorRT | 17.58 |
| YOLOv8n | PyTorch CUDA | 23.92 |
| YOLOv8n | TensorRT | 19.88 |