English | [简体中文](readme_CN.md)
# Overall process
## 1.Pytorch->TensorRT
```shell
python export.py --weights "torch's path" --onnx2trt --fp16_trt
```
## 2.TensorRT inference
```shell
python torch2trt/main.py --trt_path "trt's path"
```
Image preprocessing -> TensorRT inference -> visualization
# Time-consuming comparison
| Backbone |Pytorch(ms) |TensorRT_FP16(ms) |
|:---:|:----:|:----:|
|yolov5n-0.5| 7.7 | 2.1 |
|yolov5n-face| 7.7 | 2.4 |
|yolov5s-face| 5.6 | 2.2 |
|yolov5m-face| 9.9 | 3.3 |
|yolov5l-face| 15.9 | 4.5 |
> Pytorch=1.10.0+cu102 TensorRT=8.2.0.6 Hardware=rtx2080ti
```shell
python torch2trt/speed.py --torch_path "torch's path" --trt_path "trt's path"
```
# Visualization
<table>
<tr>
<th>yolov5n-0.5</th>
<th>yolov5n-face</th>
</tr>
<tr>
<td><img src="./imgs/yolov5n-0.5.jpg" /></td>
<td><img src="./imgs/yolov5n-face.jpg" /></td>
</tr>
</table>
<table>
<tr>
<th>yolov5s-face</th>
<th>yolov5m-face</th>
<th>yolov5l-face</th>
</tr>
<tr>
<td><img src="./imgs/yolov5s-face.jpg" /></td>
<td><img src="./imgs/yolov5m-face.jpg" /></td>
<td><img src="./imgs/yolov5l-face.jpg" /></td>
</tr>
</table>
ai_基于yolov5实现的人脸检测算法_项目实战.zip （96个子文件） 
ai_基于yolov5实现的人脸检测算法_项目实战 
detect_face.py 9KB torch2trt readme_CN.md 1KB speed.py 2KB main.py 4KB
sample.jpg 85KB readme.md 1KB trt_model.py 4KB imgs yolov5n-0.5.jpg 177KB yolov5m-face.jpg 177KB yolov5l-face.jpg 176KB yolov5s-face.jpg 176KB yolov5n-face.jpg 177KB weights download_weights.sh 278B data widerface.yaml 665B train2yolo.py 7KB coco128.yaml 2KB argoverse_hd.yaml 849B retinaface2yolo.py 6KB val2yolo_for_test.py 2KB hyp.finetune.yaml 846B coco.yaml 2KB val2yolo.py 3KB images test.jpg 416KB
Yolo5face.png 830KB pr-curves2.png 525KB result.jpg 1.21MB yolo5-top.png 184KB stem.png 12KB zidane.jpg 165KB 1.jpg 247KB yolo5.png 119KB FDDB.png 84KB bus.jpg 476KB Widerface.jpg 221KB landmark.png 1.89MB voc.yaml 738B hyp.scratch.yaml 2KB scripts get_argoverse_hd.sh 2KB get_voc.sh 4KB get_coco.sh 963B export.py 6KB hubconf.py 5KB utils __init__.py 0B google_utils.py 5KB loss.py 13KB metrics.py 8KB aws __init__.py 0B userdata.sh 1KB mime.sh 780B resume.py 1KB autoanchor.py 7KB general.py 26KB wandb_logging __init__.py 0B log_dataset.py 819B wandb_utils.py 16KB activations.py 2KB face_datasets.py 38KB google_app_engine Dockerfile 821B app.yaml 173B
additional_requirements.txt 105B plots.py 17KB datasets.py 41KB torch_utils.py 12KB infer_utils.py 1KB widerface_evaluate evaluation.py 9KB bbox.cpython-37m-x86_64-linux-gnu.so 153KB ground_truth wider_easy_val.mat 399KB wider_face_val.mat 388KB wider_hard_val.mat 414KB wider_medium_val.mat 403KB box_overlaps.pyx 2KB setup.py 328B bbox.cpython-38-x86_64-linux-gnu.so 163KB build temp.linux-x86_64-3.8 box_overlaps.o 364KB temp.linux-x86_64-3.7 box_overlaps.o 332KB box_overlaps.c 305KB README.md 507B test_widerface.py 8KB models __init__.py 0B yolov5n-0.5.yaml 1KB yolov5s6.yaml 2KB yolov5n6.yaml 2KB yolov5m.yaml 1KB yolov5s.yaml 1KB yolov5l.yaml 1KB blazeface_fpn.yaml 1KB common.py 19KB yolov5l6.yaml 2KB yolov5m6.yaml 2KB experimental.py 5KB blazeface.yaml 919B yolov5n.yaml 1KB yolo.py 16KB train.py 30KB test.py 16KB README.md 1010B资源评论
