YOLO object detection confused when see symmetrically pics

Question

I train some traffic signs that are symmetrically similar, then when i check signs, yolo is confused sometimes turn right sign detect as turn left. How can I resolve this problem. Is it related with unbalance dataset or another thing?

Answer 1

The general reasons for your error case will be

1. Data augmention. Remove the augmention step for your dataset. Re-train the model with raw data

2. Try to have as much as data possible for training. As per official docs, you must have 1000 images per instance in-order to have an average performing model. And also, be careful in not leading it to overfitting.