Can I use tfc.entropy_models to estimate the bits for fixed model? #150
-
|
Hello everyone! |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
I'm not sure what you are asking. If you already have an encoder, why not just encode the image and measure the number of bits of the encoded image? What you could do is train a prior distribution on an existing latent representation (for example the DCT coefficients in JPEG) and then use the TFC entropy model class to estimate the bits/bpp after quantization. But that would ignore/replace the entropy coding scheme that is defined by JPEG. |
Beta Was this translation helpful? Give feedback.
I'm not sure what you are asking. If you already have an encoder, why not just encode the image and measure the number of bits of the encoded image?
What you could do is train a prior distribution on an existing latent representation (for example the DCT coefficients in JPEG) and then use the TFC entropy model class to estimate the bits/bpp after quantization. But that would ignore/replace the entropy coding scheme that is defined by JPEG.