Ive been experimenting with object detection in images on my Pi5 using OpenCV. Ive found the Caffe Mobile SSD model from here https://github.com/chuanqi305/MobileNet-SSD performs the best in terms of CPU usage however i would like to improve its accuracy.

Is there any guides on how to start with retraining? I cannot find anything and the instructions on GitHub make no sense to someone new to AI models.

Thanks

Ive discovered I can annotate images using the website cvat.ai and then export the data in PASCSL VOC 1 format which I think is correct for Mobile SSD caffe models? Any advice from there would be great thanks

Hi! MobileNet-SSD is fast but limited in accuracy for new objects. To improve it:

Collect & label images – use LabelImg
in VOC format.
Set up training – better on a PC with GPU, install Caffe.
Modify model – change num_classes in .prototxt to match your objects.
Fine-tune – start from pre-trained weights (MobileNetSSD_deploy.caffemodel) and train with your data.
Test – load new model in OpenCV using cv2.dnn.readNetFromCaffe().
Tip: start small (2–3 classes) before scaling up.