Mike

> But now I do not think introducing auxiliary targets like 'image_cache' and 'image1' is a good approach Yes, I read you comment in your MR that is why I...

I see, yes, makes sense to have `Compose` for multi-input / single-output images. > I have been thinking of this issue for the past few days and have just started...

I will pick this up again if the `multi-input / single-output` PR gets merged. Not worth investing the time of adapting this, if it is not happening

> First, basically tried to apply only my video source https://github.com/mikel-brostrom/yolo_tracking/issues/1263#issuecomment-1898657606 > Second, tried to apply my video + pre-trained yolov5s model Yolov5 is not official supported in this repo...

Values larger than `cost_limit` are disregarded. Which means that there will be unmatched rows.

Is this still happening @WongKinYiu ?

Wow, this looks fantastic @i-aki-y. Should I adapt MixUp in my PR and make it inherit `BatchBasedTransform` if this is the new way of working with multi-image augmentations @Dipet?

Any plans of getting this merged any time soon?

Thx for the link @Marco2929! The problem with this package is that it is full of relative imports causing collision with local files. `import utils...` collides with `utils.py` for example

Maybe an argument to specify 1/3 channels to return would be enough? ```python def to_gray(img, gray2rgb=False): gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY) if gray2rgb: return cv2.cvtColor(gray, cv2.COLOR_GRAY2RGB) else: return gray ```