English | 中文

Visual Detection Interface

Python interface for real-time visual recognition using a camera

Overview

This package provides: Real-time recognition for cameras

Environment Requirements

-Linux/Windows environment -Python 3.7+ -PyTorch>=1.7

Installation and Running Process

1. Download or clone the repository: git clone git@192.168.10.116:vision/tridexhand_vision.git # Clone

2. [Note: If you don't have Anaconda3, you can skip this step] Create a virtual environment (using Python 3.11 as an example): conda create -n py11 python=3.11

3. Install the necessary packages: pip install -r requirements.txt # Install

4. Edit the config/config.yaml configuration file to set parameters.

5. Run the test code: python detect.py

Parameter Description

To perform inference with your own model and dataset, you need to adjust the following key parameters in detect.py according to your needs and characteristics of the dataset:

Model Weights:

Specify the path to your custom model using the weights parameter. Default: weights best.pt

Data Source:

Select the input source with the source parameter, which can be a file, directory, URL, or camera. Default: Using a camera: source 0

Inference Image Size:

Set the input image size for inference with the imgsz parameter. Default: imgsz 640

Confidence and IoU Thresholds:

Adjust conf-thres and iou-thres to control the confidence and Non-Maximum Suppression thresholds. Default: Set confidence threshold to 0.3: conf-thres 0.3

Device Selection:

Specify the use of GPU or CPU with the device parameter. Default: Use a specific GPU: device 0

save_txt: true

Description: Controls whether to save detection results to a text file. When this option is enabled, information about each detected object (such as class and position) is saved in text format.

save_crop: true

Description: Controls whether to save each detected object's region as a separate image file after detection.

save_conf: false

Description: Controls whether to include confidence scores in the saved label files. If set to false, confidence values will not be saved.

view_img: false

Description: Determines whether to display the inference results in a window. If set to true, a window will open to show the detection results; if false, it will not display.

classes: null

Description: This is a filtering option. It allows you to specify indices of classes of interest (such as [0, 1]) to filter output results. If null, it means no filtering is applied.

project: "runs/detect"

Description: The main directory location for saving results. All detection results will be saved here.

name: "exp"

Description: The name of the subdirectory for saving results. A subdirectory is created under the main project directory to save results from each detection run.

line_thickness: 3

Description: The thickness of the bounding box lines, in pixels. This affects the prominence of the detection boxes in the output image.

vid_stride: 1

Description: The frame stride for video processing. A value of 1 means processing every frame. To speed up processing, you can set it to a larger value like 2 or 3.