Object Detection Comparison

Benchmarks and compares six object detection models on image datasets and video:

• YOLOv8
• RT-DETR
• Faster R-CNN
• RetinaNet
• SSDLite
• FCOS

Metrics evaluated: Precision, Recall, and Mean Average Precision (mAP).

Setup

python -m venv object_detect_env
source object_detect_env/bin/activate
pip install ultralytics torchvision torchmetrics opencv-python pillow

Model Weights

Model weights are not included in this repo. Download them before running:

YOLOv8m (yolov8m.pt):

python -c "from ultralytics import YOLO; YOLO('yolov8m.pt')"

RT-DETR-L (rtdetr-l.pt):

python -c "from ultralytics import RTDETR; RTDETR('rtdetr-l.pt')"

Both files will be downloaded automatically to the current directory on first run. Faster R-CNN, RetinaNet, SSDLite, and FCOS weights are downloaded automatically by torchvision.

Dataset

Evaluation uses the COCO val2017 dataset. Download it and update the paths in object-detect.py before running.

Usage

python object-detect.py