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
Follow setup instructions readme.
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.
- In
data_generation_parameters.jsonspecify"total_frames":100and"background":"blank". python generate_data.py
- Use
create_pose_json.pyto generate apose_list.jsonfile which will contain the[azimuth, elevation, tilt, scale]per frame of the data you would like to render. - Example command to render 10 objects according to pose specified in
pose_list.jsonpython generate_data_pose_list.py -start=0 -end=10
- If using GPUs to render, specify bigger a render tile size in
data_generation_parameters.jsonto speed up rendering, and similarly a smaller one if using CPU.
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}
}









