Giters

shunsukesaito / PIFu

This repository contains the code for the paper "PIFu: Pixel-Aligned Implicit Function for High-Resolution Clothed Human Digitization"

Home Page:https://shunsukesaito.github.io/PIFu/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Training on my own data

SeanChenxy opened this issue · comments

commented

Hi, @shunsukesaito. Thanks for sharing this code. I want to use PIFu in my own dataset, so I wonder whether some setting should be changed. For example:

  1. To collect my own OBJ file, is there any suggestion on the unit (m/cm/mm)?
  2. If I scan a new people OBJ file by myself, which point should be the coordinate origin ( i.e., (0,0,0))?
  3. Could you suggest how to set self.B_MIN and self.B_MAX (in class TrainDataset) according to my own dataset?
commented
  1. The repo uses cm.
  2. Take a look at the data provided as example. The people are centered in x and z, and originate at y=0. So the coordinate origin is the center between the feet.
  3. Set the bounding boxes to where it makes sense. Eg. calculate the min/max extent of the axis aligned bounding boxes of your data and add a couple of cm (eg. 20 cm) as "safety margin".