jonarchist
commented
2 years ago Yes, that's correct. bits is the number of bits required to encode one image. Bits per pixel are computed in this line.
.tfci is just a file extension we chose to write the compressed images to. You can write/read them with the same Python script.
The compressed stream is returned by the compress method in this line. It's the first return value, followed by the image and latent shapes.
Hello Dr. Balle, I have tried to use bls2017.py to train my own model, but I'm not sure that whether bpp is estimated or not when training the model. So I printed the "bits" in "y_hat, bits = entropy_model(y, training=training) line 115", but found the numeric value (like 17400.3, I guess it is the number of bits required for encoding an image). Also, I want to know what is the .tfci file and how can I get the stream of transmitted bits after a picture is compressed, please tell me how to do when you're available, thanks!