I made a project which detected the severity of car damage and classified it within 5 different classes. I’ve a dataset of 3471 images (including augmentation) of different resolution from 640x480 to 1024x768 (with bounding box annotations).
I followed this guide step-by-step to build my project.
I trained the data for 125 epochs with epoch_length=100 (fairly low, but it took nearly 12 hours with my GPU).
At the testing phase of the model, I see no output (no bounding box predictions) on the test images (was expecting results with less accuracy). Thinking low epochs must be a reason, but even after the adjustments, result was the same.
I re-checked my code, CSV files, bounding box coordinates, plotted some of the bboxes of images to check they were correct, but didn’t figure out anything wrong.
Any help regarding it would be appreciated…