This is a repository for the altered YOLOv5 codebase and scripts used in the Master Thesis - Improved Sheep Detection, written in the spring of 2022. The altered YOLOv5 model found in this repository makes it possible to run YOLOv5 with a fourth channel of information, in our case this was Infrared Radiation (IR) imagery.
The branch master only contains the standard model to process RGB images.
The branch four-input contains the model which combines RGB images and IR images as 4 channels before processing the image.
The branch generic-fusion contains the model which processes RGB images and IR images separetly in the backbone, and combines the results in a fusion module before the neck/head.
The data folder is ignored, but should look like this:
- data/
- train/
- images/*.jpg
- ir/*.jpg
- labels/*.txt
- validation/
- images/*.jpg
- ir/*.jpg
- labels/*.txt
- test/
- images/*.jpg
- ir/*.jpg
- labels/*.txt
- train/