Training fails with RF-DETR inst - Input size should be a multiple of 24

[leoll2]

Steps to reproduce

1. Open the demo segmentation project (horses)
2. Train a model with RF-DETR-Seg-S

Training fails immediately with the following error:

  File "/home/leoll2/training_extensions/library/src/otx/backend/native/models/base.py", line 170, in __init__
    self.data_input_params = self._configure_preprocessing_params(data_input_params)
    │                        │    │                               └ DataInputParams(input_size=(120, 211), mean=(0.0, 0.0, 0.0), std=(1.0, 1.0, 1.0))
    │                        │    └ <function OTXModel._configure_preprocessing_params at 0x7499061c14e0>
    │                        └ RFDETRInst()
    └ RFDETRInst()

  File "/home/leoll2/training_extensions/library/src/otx/backend/native/models/base.py", line 1011, in _configure_preprocessing_params
    raise ValueError(msg)
                     └ 'Input size should be a multiple of 24, but got (120, 211) instead.'

ValueError: Input size should be a multiple of 24, but got (120, 211) instead.

Full log: bug-rfdetr.log