Skillnoob

you can use the [`to_json()`](https://docs.ultralytics.com/reference/engine/results/#ultralytics.engine.results.Results.to_json) method on the [`Results`](https://docs.ultralytics.com/reference/engine/results/#ultralytics.engine.results.Results) object that gets returned.

Also, the [auto_annotate](https://github.com/ultralytics/ultralytics/blob/53e5d02a289ae9440dde019e90af88481120dea0/ultralytics/data/annotator.py#L8) function doesn't return anything, so you'd need to modify the source code to fit your needs: https://github.com/ultralytics/ultralytics/blob/main/ultralytics/data/annotator.py

See the [docs](https://docs.ultralytics.com/modes/predict/#inference-arguments). `verbose=False`

Agree with this change, there were multiple issues in the past with the same issue and this behavior makes more sense.

You can can add background images that don't contain any of the objects you are trying to detect to your dataset. These images don't need annotation files. You should add...

@tjasmin111 https://docs.ultralytics.com/guides/model-training-tips// This guide contains some general guidance for making a good dataset, ignore the parts about yolov5

@Petros626 ```python from ultralytics import YOLO model = YOLO('yolo11n.pt') model.val(data='dota8.yaml', batch=1, visualize=True) ``` You also find this in the [val docs](https://docs.ultralytics.com/modes/val/).

Are your training images at least 1280x1280? If they aren't, it will need to upscale the images which will give you no benefit.

Can you share the results like @Y-T-G asked for.

@MasIgor for those kinds of images I would reccomend [SAHI](https://docs.ultralytics.com/guides/sahi-tiled-inference/) and training with a lower imgsz like 640 or 1280, because training with a imgsz of 4000+ would take too...