CRIB Data Generator

Training Data - Learning Exposure

Testing Data - Random Views

This repository contains a set of Python and Blender scripts for data generation in Incremental Object Learning from Contiguous Views

Initial Setup

Follow setup instructions readme.

Quick Demo Commands

python generate_data.py -start=0 -end=1 

Generates one small learning exposure in train_data and testing samples in test_data of one object on a blank background.

python generate_data_pose_list.py -start=0 -end=1 

Make sure to execute these commands in the main repository directory, not ./CRIB. Generates 10 frames of one object rotating in the tilt direction (euler coordinates), specified in pose_list.json.

Generating Data as in the Paper

1. In data_generation_parameters.json specify "total_frames":100 and "background":"blank" or "background":"clutter" depending on background preference.
2. python generate_data.py

Generating Data From Specified Pose

1. Use create_pose_json.py to generate a pose_list.json file which will contain the [azimuth, elevation, tilt, scale] per frame of the data you would like to render.
2. Example command to render 10 objects according to pose specified in pose_list.json python generate_data_pose_list.py -start=0 -end=10

Additional Notes

1. If using GPUs to render, specify bigger a render tile size in data_generation_parameters.json to speed up rendering, and similarly a smaller one if using CPU.

Citation

If you use this code, please cite our work :

@InProceedings{Stojanov_2019_CVPR,
author = {Stojanov, Stefan and Mishra, Samarth and Anh Thai, Ngoc and Dhanda, Nikhil and Humayun, Ahmad and Yu, Chen and Smith, Linda B. and Rehg, James M.},
title = {Incremental Object Learning From Contiguous Views},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2019}
}