Skip to content

fbh16/small_object_detection

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Slice training method for YOLOv5

This repository is dedicated to the YOLOv5 model for small target detection. Its main objective is to improve the model's accuracy in detecting small objects by training it on sliced images.

demo

Dataset

This repository utilizes the VisDrone dataset (https://github.com/VisDrone/VisDrone-Dataset.git) for slice preprocessing and trains a YOLOv5 model with the sliced dataset.

The detection performance of the model is shown in the following images:

demo

How to run

1. Prepare the dataset according to the format in "EXAMPLE" under the link: https://roboflow.com/formats/yolov5-pytorch-txt?ref=ultralytics, and store the dataset images and labels in the images and labels folders respectively.

2. Modify the parameters in slice.py such as root, slice_width, slice_height, overlap_h_ratio, classes, etc., to suit your needs.

3. Run slice.py

Parameters

slice_width and slice_height: Represent the size of the slice boxes, overlap ratio is used to avoid the situation where bounding boxes are being devided.

overlap_h_ratio and overlap_w_ratio: Move a certain distance back in the original image's height and width directions respectively after each slicing, thereby preserve the training value of the segmented bounding boxes.

val: Used for validating the correctness of the slicing results. When set to True, it will visualize the bounding boxes of the first 'n' sliced images (where 'n' is a specified number).

classes: Target categories contained in the dataset.

Get the results

The results of the slicing are saved in the " ../runs/exp" directory.

The slice_images folder is used to store the sliced images.

The slice_labels folder is used to store the sliced labels.

The validate folder contains visualization images of the slicing effects.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages