YOLOv7 Image and Video Inference with UI

This project is made for having quick inferences and makes the job easier for end users. We aimed to develop a light project that makes fast inferences.

As an example, we used pothole dataset.

The system features:

OS : Ubuntu 20.04 LTS 64-bit 
CPU : Intel(R) Core(TM) i5-10200H CPU @ 2.40GHz
GPU : Nvidia GeForce GTX 1650ti 4GB
RAM : Samsung M471A1K43DB1-CWE 8GB

As you can see above, this project can also work on mid-segment systems.

Inferences on Photo and Video

Cloning the Repository

Clone this repository with git.

  git clone https://github.com/oguzaybilir/YOLOv7-Predict-with-UI.git
  cd YOLOv7-Predict-with-UI

Installing Libs

There is a requirements.txt file to install packages you need. This file contains almost all libraries and modules used in the project.

To install this libraries and packages:

    pip3 install -r requirements.txt

Required Packages

These packages are absolutely essential packages for this project. In this case, you must first install the following packages in this order.

The nvidia-driver-xxx is your driver which is compatible with your graphic card.

    nvidia-driver-xxx
    CUDA == 11.6.2
    torch == 1.12.1
    TensorRT == 8.4

TRT Weights and Sources

The converted trained weights and sources are stored in the drive link below.

Drive Link to .trt weights and sources.

Converting .PT weights to .ONNX and .TRT weights

Method 1

Pytorch to ONNX with NMS (and inference) - Open In Colab

python export.py --weights yolov7-tiny.pt --grid --end2end --simplify \
        --topk-all 100 --iou-thres 0.65 --conf-thres 0.35 --img-size 640 640 --max-wh 640

Method 2

Pytorch to TensorRT with NMS (and inference) - Open In Colab

wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-tiny.pt
python export.py --weights ./yolov7-tiny.pt --grid --end2end --simplify --topk-all 100 --iou-thres 0.65 --conf-thres 0.35 --img-size 640 640

git clone https://github.com/Linaom1214/tensorrt-python.git
python ./tensorrt-python/export.py -o yolov7-tiny.onnx -e yolov7-tiny-nms.trt -p fp16

Method 3

Pytorch to TensorRT another way - Open In Colab

wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-tiny.pt
python export.py --weights yolov7-tiny.pt --grid --include-nms

git clone https://github.com/Linaom1214/tensorrt-python.git
python ./tensorrt-python/export.py -o yolov7-tiny.onnx -e yolov7-tiny-nms.trt -p fp16

/usr/src/tensorrt/bin/trtexec --onnx=yolov7-tiny.onnx --saveEngine=yolov7-tiny-nms.trt --fp16

Run

  python3 main.py

Authors

• @oguzaybilir
• @furkantahabademci

Acknowledgements

• Mehmet Okuyar
• Official YOLOv7 Repository
• TensorRT Website
• CUDA Website