Laughing

@Racheal-c I'm closing this issue as the same issue had been solved. Feel free to reopen if it still happens after you check the data. :)

@mrmarten hi actually I can't reproduce your issue with the same code: ```python from ultralytics import YOLO model = YOLO("weights/yolov8n.pt") results = model.predict(source="./ultralytics/assets/bus.jpg", visualize=640, show=True) ``` it just showed once...

@jalexhurt the save_dir is composed of `project` and `name`, which means `save_dir = project/name`, so just modify these two if you want to change save_dir.

@quangsonle hi detection is detection itself, classification is classification itself. So it you're going to use detection then please go ahead with yolov8. :)

@dmddmd also recommend yolov8 if you want to use yolo model as backbone. :)

@QwerMotion @knoppmyth hi guys, the PRs are merged, please using the latest version then the usage of PIL/numpy should work. ```bash pip install --upgrade ultralytics ```

@aalling93 hi, yeah we need images as original training sources. You can add `cache=disk` amd it'll automatically generate `.npy` files in your images folder then the training will use the...

@ahthserhsluk hi, may I ask if there're corrupt labels in your datasets?  actually I just found that we got a mismatch issue if there's corrupt label in datasets, so...

@ahthserhsluk @fcakyon @luigisaetta hi guys, the nan loss issue could be solved in this PR #490. We'll merge it later today. :)

@guishilike if you're training detection and want to modify `mosaic`, probably you can add `v5loader=True`. the new loader haven't supported modifying `mosaic` ratio, we'll work on it. :)