Environment

• Ubuntu 24.04
• Python 3.10

SETUP

put SUNRGBD dataset folder in metadata folder

SUNRGBD dataset download link: https://rgbd.cs.princeton.edu/data/SUNRGBD.zip

it should look like this:

Extract-SunRGBD-Data/
└── metadata/
    └── SUNRGBD/
        ├── kv1/
        ├── kv2/
        ├── realsense/
        └── xtion/

# Clone repo 
git clone https://github.com/MRamazan/Extract-SunRGBD-Data

# Setup
cd Extract-SunRGBD-Data
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Extract data
# You can set classes in extract_sunrgbd_data.py file
python3 extract_sunrgbd_data.py --save_folder "sunrgbd_labels" --save True --save_imgs True --save_pcd True --save_votes False --sample_point_count 100000

PCD

Point Cloud Data (pcd)

The point cloud data is represented as an array with the following structure:

• Shape: [N, 6] where N is the number of points.
• Each point consists of:
  • 3D Coordinates (x, y, z): The first three columns (pcd[:, 0:3]).
  • RGB Color (r, g, b): The last three columns (pcd[:, 3:6]).

#pcd: [N,6] point cloud data created from depth image 
#np.array([N, 6]) N = number of points 
pcd = np.load("pcd/000001.npz") 
points = pcd[:, 0:3] 
colors = pcd[:, 3:6]

Calib

Calibration Matrix (calib)

The calibration matrix consists of two parts:

• Rotation Matrix (Rtilt): Defines the orientation of the object.
• Intrinsic Matrix (K): Contains the camera's intrinsic parameters.

calib: {
  Rtilt: [
    [x, x, x],
    [x, x, x],
    [x, x, x]
  ],
  K: [
    [fx, 0, cx],
    [0, fy, cy],
    [0, 0, 1]
  ]
}


with open("calib/000001.txt", "r") as file:
     calib = json.load(file)
     Rtilt = calib["Rtilt"]
     K = calib["K"]
#np.load("votes/000001.npz")

VISUALIZE

#export XDG_SESSION_TYPE=x11
python3 visualize.py --image_index 11 --visualize_function 1

#uncomment the line below if you get window error
#export XDG_SESSION_TYPE=x11
python3 visualize.py --image_index 11 --visualize_function 2