ObjectDrawer-ToolBox

ObjectDrawer-ToolBox is a graphical image annotation tool to generate ground plane masks for a 3D object reconstruction system, Object Drawer.

Prerequisites

• Python 3.8
• ffmpeg

Requirements

opencv-python
ffmpeg-python==0.2.0

Getting Started

Given the video captured by conventional recording devices, the ObjectDrawer-ToolBox is used to sample frames from the video and label the ground plane masks. In the ground plane mask task, you need to label 3 images and only keep ground pixels in these images.

Please run

python label_image.py \
  --video_path /path/to/your/videos.mp4 

How to annotate ground plane

1. Draw a polygon to cover the pixels area which are not belong to groud plane. Tips: red / black line denotes a unfinished / finished polygon.
2. Press w to delete non-ground pixels when finished a polygon drawing.
3. Repeat the step 1 & 2, until there are only ground plane areas in the image.

Operation Tips:

• Press w to delete pixels.
• Press d to finish current annotation and start to label next image.
• Press esc to clear candidate regions

select pixels

delete pixels

Annotation result

After ground plane annotation, a zip file named “label_${videos}.zip” is generated in same directory as the input video.
Upload the video (“videos.mp4”) and zip file on the website of Object Drawer.

Examples

To clarify the label data format, we provided examples video & label file. You can download these data and submit to the Object Drawer website for testing.

Acknowledgements

Thanks to PlenOctree for the octree converter and online viewer. We take U2NeT as the segmentation algorithm. Please consider citing their papers and following their license.

GitHub

View Github