[Bug] Hardcoded Input Shape (1,3,224,224) in PyTorch Quantization

[MadhurKumar004]

This assumes every model is a standard ImageNet computer vision model. Attempting to quantize audio models, NLP models, or even MNIST models (1 channel vs 3) will cause an immediate crash during the calibration step.

[Manav13254]

so if i understand this issue correctly it has hardcoded value for the input to be an image always and the requirement is to accept a dynamic input shape ?

[MadhurKumar004]

@Manav13254 Yes