Logic for resizing images to fit the size of the model input data.

[oneprofittips]

Add or change in parameter
Mode = originalAspectRatio ? ResizeMode.Max : ResizeMode.Stretch
method for resizing an image, originalAspectRatio suggests leaving the original size, but in a fluid situation, if the size is 480 x 60, for example, with a model size of 480 x 480, it will stretch the smallest side to fit the size of the model.
Instead of ResizeMode.Max, I think it’s better to change it to ResizeMode.BoxPad or display these settings in the configuration.

[dme-compunet]

Sorry for the late reply, what are you trying to get from BoxPad? It is exactly like the current behavior, in the current state the input tensor is created in the size of the model, the pixels of the image are copied to the center of the tensor and the margins remain with a value of 0.0 which means black color. (BoxPad will only add time of copying the black pixels of the padding)

[oneprofittips]

For example, I do training on a size of 480x480 and pass 480x80 to the input of the Detect model, and I get 0 matches.
If I train at 480 and pass 480x80 to the input during training, and with Detect 480x80 I get a more or less good result.
It turns out that the resize is stretching 80 pixels by 480 pixels, and not adding empty pixels.

[dme-compunet]

I don't know how to solve your problem, but I don't think BoxPad will help, if YoloV8Parameters.KeepOriginalAspectRatio is set to true, the aspect ratio is kept as in the original. Try to fork this repo and see if changing ResizeMode.Max to ResizeMode.BoxPad will solve your problem, if it solves your problem, please share here.

[oneprofittips]

You can close, I rewrote the application in python.
Without batch input, resources are wasted.