About get_scale_table(min=SCALES_MIN, max=SCALES_MAX, levels=SCALES_LEVELS)

[Junx0924]

Documentation

The original default values are:

https://github.com/InterDigitalInc/CompressAI/blob/release/1.1.1/compressai/models/priors.py

SCALES_MIN = 0.11 SCALES_MAX = 256 SCALES_LEVELS = 64

When the input image data is 16 bit instead of 8 bit, should these values be changed according to the data range? and how to decide the SCALES_LEVELS with new data range?

[jbegaint]

Probably not. Though you can check easily by (1) passing your own scale_table to the update function here and (2) evaluating the model performances on some images.

[YuKDseele]

Hi, I wanted to ask if you have resolved this problem, as I am now encountering a similar problem. I am using 12-bit images and unsure whether I need to modify SCALES_MAX and SCALES_LEVELS. However, from the results, it seems that compression can still work, although the results may not be ideal.