Have you ever encountered a strange issue with Label Studio where your YOLO outputs just wouldn’t make predictions, but then you switch to v8 OBB format and suddenly everything works like a charm? I recently stumbled upon a Reddit post where someone was facing this exact problem.
The user was trying to create a cat recognition algorithm and had uploaded 50 cat photos to Label Studio. They exported the labeled data in YOLO format, trained a model with v11, but to their surprise, the model couldn’t produce any output, not even on the training photos themselves.
But here’s the interesting part: when they exported the same data in YOLOv8 OBB format and trained it, the model achieved a recognition rate of 0.97! So, what was going on?
I’m going to take a stab at explaining this mystery, but I’d love to hear your thoughts in the comments too. Is there something I’m missing?
One possibility is that there’s an issue with the way Label Studio is exporting the YOLO data. Maybe there’s a formatting problem or some other quirk that’s preventing the model from making predictions. On the other hand, it’s also possible that there’s a bug in the v11 model itself that’s causing the problem.
Whatever the reason, it’s clear that switching to v8 OBB format solved the issue, but it’s still puzzling why YOLO outputs didn’t work in the first place. Have you encountered similar problems with Label Studio or YOLO outputs? Share your experiences in the comments below!