In C++, calling the ONNX model to process images only yields the lower half of the data.

[mik4396]

With the same code and ORT environment, other models work fine, but with ben2, only the lower half is displayed.

[mik4396]

Currently, I'm working by adding empty data of the same height above the image, then removing this data after inference. Incidentally, my ORT version is 1.23.2. Using other versions will result in different phenomena; for example, ORT version 1.21 also only shows the lower half of the image and the image is shifted to the right.