Potential Precision Loss When Using Normal Data in PNG Format

[wzy-99]

I would like to bring attention to a potential issue regarding the precision of normal data in the get_normal_filepaths method.

dn-splatter/dn_splatter/data/coolermap_dataparser.py

Line 99 in aa754d8

return natsorted(glob.glob(f"{self.normal_save_dir}/*.png"))

Currently, normal data is being stored in .png and .npy format. But this method only uses png, which may result in precision loss.

[maturk]

That is an interesting point ... I have not considered this.