search

shunsukesaito / PIFu

This repository contains the code for the paper "PIFu: Pixel-Aligned Implicit Function for High-Resolution Clothed Human Digitization"
https://shunsukesaito.github.io/PIFu/
Other
1.75k stars 343 forks source link

Inconsistency between TrainDataset and EvalDataset's code #128

Open jsalzRR opened 2 years ago

jsalzRR commented 2 years ago

Hi, I happened to stumble upon this : In TrainDataset.get_render :

            # Match camera space to image pixel space
            scale_intrinsic = np.identity(4)
            scale_intrinsic[0, 0] = scale / ortho_ratio
            scale_intrinsic[1, 1] = -scale / ortho_ratio
            scale_intrinsic[2, 2] = scale / ortho_ratio

In EvalDataset.get_render :

            # Match camera space to image pixel space
            scale_intrinsic = np.identity(4)
            scale_intrinsic[0, 0] = scale / ortho_ratio
            scale_intrinsic[1, 1] = -scale / ortho_ratio
            scale_intrinsic[2, 2] = -scale / ortho_ratio

Is the fact that there is an additional minus sign in EvalDataset wanted ? If it is, I would be grateful if you could explain why these implementations are different :)

Good day !