AI Model Efficiency Toolkit (AIMET) Forum

InputShape in SSVD API

I have installed AIMET on google colab. I am trying to evaluate AIMET using Pytorch APIs. I have a question regarding input_shapes parameter in compression api. Our dataset has input images of different sizes. We are using MaskRCNN model for segmentation which supports input images of different sizes. In that case what value needs to be passed as input_shape?

(Internally, each input image is transformed into an image of fixed size in initial layers. However the size (after transformation) varies based on the actual image data in the batch)

Good question @amitdedhia.

The input_shape parameter is used to create random tensors that we can use to pass some data through the model. If for your particular model, there is a choice of valid input_shapes, you can pass any one of them.


Thank you for the response