Specify max_scale in tf.classify

For one of our applications, we need to detect small objects that can occur in multiple subsets of the image, i.e., we know that the detected object will not exceed a particular size and it will never fit the entire window/ROI.

Is there a way to specify the max_scale of the tf.classification so that it does not run the function on the entire image and starts from, say, a scale of 0.5 and goes down to a scale of 0.25? Currently the documentation says the function assumes a default value of 1:

The sliding window method works by multiplying a default scale of 1 by scale_mul while the result is over min_scale .

Currently no, it’s a pure image pyramid so I assume you’d want the largest picture always.