haosulab / ActiveZero

[CVPR 22'] ActiveZero: Mixed Domain Learning for Active Stereovision with Zero Annotation
https://www.liuisabella.com/ActiveZero/
8 stars 0 forks source link

paper #1

Closed AI-student-wl closed 1 year ago

AI-student-wl commented 1 year ago

Hello, thank you for your excellent work. I want to reproduce this work, and I want to know some details of your data set production and the placement of data files: RGB, Depth, 2 IR image, sensor internal parameters. Whether sensor pose is required, it is a 4 4 matrix consisting of rotation matrix R and translation matrix T. I use d435i to make data sets. Thank you for your help

Isabella98Liu commented 1 year ago

Hi, thanks for your question. The sensor pose is required and yes, it's a 4*4 pose matrix. In this project we use Intel RealSense D415 to build the real dataset.

The simulation data files contain: 2 IR images w/ infrared patterns, 2 IR images w/o infrared patterns, 2 * depth images, camera meta files.

The real data files contain: 2 IR images w/ infrared patterns for different infrared pattern strength (we use 6 different IR intensities, so in total 26=12 IR images), 2 depth images, camera meta files.

AI-student-wl commented 1 year ago

How do I collect the left and right infrared images with different intensities? I have only one intensity here. If it is code acquisition, can I see your acquisition code? Thank you for your help

AI-student-wl commented 1 year ago

img_real_L, img_real_L_no_ir, What is the difference between the two data?

AI-student-wl commented 1 year ago

In the messytable.py file, img disp L and img real L temporal What does ir mean?