bytedance / BEE

Apache License 2.0
21 stars 8 forks source link

KeyError: 'gaussian_conditional._quantized_cdf #2

Closed DL19970112 closed 2 years ago

DL19970112 commented 2 years ago

Hello, After downloading the pretrained models, run: "python3 Encoder/CoreEncApp.py -i kodim01.png -o kodim01.bin --ckpt checkpoints/model.ckpt-02 --target_rate 0.06" error: "KeyError: 'gaussian_conditional._quantized_cdf'"

Could this be due to compilation problems in Compressai ? How to solve this problem?

DL19970112 commented 2 years ago

The environment I used is "pytorch1.9.0, python3.8, cuda11.4" I built compressai1.1.4 in Python3.8 and obtained _cxx.cpython-38-x86_64-linux-gnu. so and ans.cpython-38-x86_64-linux-gnu.so, And replace these two files in the BEE-main /Common folder. The ”KeyError: 'gaussian_conditional._quantized_cdf” is still there.

Wubyte commented 2 years ago

Thanks for your bug reporting! This issue is caused by the encrypted decoder, we already fixed it in the latest version v1.0.1. Pls download the latest code version and checkpoint version to fix this issue.