QRNet: Quaternion-based Refinement Network for Surface Normal Estimation

We will gradually improve the code base:

• Submit log messages
• Submit model weights and test code
• Submit training related code

Submit log messages

Due to the size of the log file, a download method will be provided [Password: 2333].

logs/NYUv2.log : Training log applied to the NYUv2 dataset.
logs/ScanNet.log : Training log applied to the ScanNet dataset.

Dataset	Detailed Accuracy	Structural Accuracy
NYUv2	64.0211	85.6046
ScanNet	73.6758	90.2324

The source code will be released upon acceptance