-
Take photos of your environment of two or more objects. (at least 100 instances between all objects)
-
Annotate them on roboflow.
-
Train a Faster RCNN model using detectron2
-
Train Yolov4/5/6/7/8 (only one of them of choice) the smallest size
-
Evaluate both models based on mAP and speed and size.
2 Objects' pictures were taken. 1. Pyramid, 2. Glasses Pyramid image :
Glasses image:
- confusion matrix
- F1 curve
- PR curve
- P curve
- R curve
- Faster RCNN: ~0.01%
- Yolov8: ~0.13%
Yolo is more speedier despite its size, and its speed has a number of advantages. Yolo training takes 3 to 4 minutes for 24 epochs, but Faster RCNN (detectron2) takes more than an hour.
- Bigger RCNN model size: 805.5 Mb
- The Yolov8 model size: 21.53 Mb
-
https://blog.roboflow.com/how-to-train-yolov8-on-a-custom-dataset/
-
https://blog.roboflow.com/how-to-train-detectron2/#using-your-own-data-with-detectron2