Laughing

@plashchynski hi, it looks like a video format specific issue. I've also tested a video using the same code as your except the video_path and the saving works fine to...

@sstainba looks there's something wrong about your PIL/ font. But this error occurs in the image-plotting process of the first epoch, actually it won't affect the training process itself, so...

@sstainba ok I'll add some empty anno files in my case to reproduce the `IndexError: boolean index did not match indexed array along dimension 0; dimension is 1 but corresponding...

@sarmientoj24 yes you can use yolov5 segmentation data format here. it should work. can you share more error log? also the command please. Thanks!

@sarmientoj24 can you use the latest version(8.0.5) and try again? just `pip install --upgrade ultralytics`. or if you can share the datasets to me then I can debug and try...

@sarmientoj24 oh actually both your errors look like image-related issue. can you share some images? or just tell me more info about your images, like resolutions or if it's gray...

@sarmientoj24 hi some users met the same `torch.stack` issue when they set `mosaic!=1.0`. ```bash value = torch.stack(value, 0) TypeError: expected Tensor as element 1 in argument 0, but got tuple...

@ercanhocaninyeri @AyushExel there are some reasons that you got slower speed with `model.predict`. - the first time you run `model.predict` it need to initialize the predictor which could cost some...

@ch-hristov the scale is a arg for affine augmentation, not the multi-scale training. @AyushExel actually I think we haven't added the functionality of multi-scale yet.

@ZongXR yes! https://github.com/ultralytics/ultralytics/blob/d021524e850acfa393ec25d4ecb9c4c761cca688/ultralytics/cfg/default.yaml#L37