YOLOv7 Fine-Tuning Outperforms: Camouflage Detection Comparison

Nowadays camouflage object detection is very challenging due to its complex background, change in shape as well as size of an object. In past, many algorithms have been designed to detect camouflage object but had problem of increase in false positive as well as facing with some failure cases. Initially Yolov5 (You Only Look Once version 5) with all its variant is tested on a COD dataset. Afterwards to improve the detection results, Yolov7 is trained and tested on same dataset. To reduce the failure cases and improve the result, Yolov7 is fine-tuned and this is also train and tested on that dataset. The experimental results show that fine-tuned Yolov7 gives increased mAP_0.5:0.95 by 5% than YOLOv7 and 7% than YOLOv5s. Ultimately fine-tuned Yolov7 gives the best mean average precision (mAP) increasing overall performance of detection than other YOLO variants.