README.md

Prepare SUN RGB-D data

1. Download SUNRGBD v1 data from HERE (SUNRGBD.zip) and the toolbox (SUNRGBDtoolbox.zip). Move all the downloaded files under data/sunrgbd. Unzip the zip files. Ensure the directory is like this.

data/sunrgbd/
├── SUNRGBD
    ├── kv1
    ├── kv2
    ├── realsense
    └── xtion
└── SUNRGBDtoolbox

2. Change the PROJECT_DIR of extract_rgbd_data.m according to your project position. Extract point clouds and annotations by running extract_rgbd_data.m under the sunrgbd/matlab folder. It will re-organize the original dataset in sunrgbd/mysunrgbd folder.

sunrgbd/mysunrgbd
└── training
    ├── calib
    ├── depth
    ├── image
    ├── label
    └── pc

3. Extract frustum data by running python sunrgbd/prepare_data.py --gen_train --gen_val --gen_val_rgb_detection

Training and evaluation

bash scripts/sunrgbd_train.sh