I’m new to object detection and machine learning. I’m working on object detection project using a PTZ camera and thermal camera. My model is trained in Google Colab with multiple Roboflow datasets (yolov8n). I have 4 models (human, animal, boat, fire), each with different color bounding boxes.
I’m facing accuracy issues where objects are not detected correctly such as humans being misidentified as animals. Does anyone have suggestions for improving performance or fine-tuning the model for better accuracy?
I just got into this as well. From my relatively short experience (less than a week) a smaller high quality dataset is better than a larger low quality dataset.
Secondly, make sure that the classes for each model are roughly equal … so if you have if you have roughly 50 classes of human_objects make sure you have roughly the same amount for every other class.
It is highly likely that your model is inaccurate because the training dataset has an unequal representation of classes.
So try build a smaller model (± 50 images/class) and see if your results don’t improve
