README

SemanticHouseCat3D is a synthetic dataset for semantic matching where each category directory contains a number of rendered frames for each objects. The typical structure of a category directory is the following:

📁 category

📁 depth # depth maps of each frame

🖼️000000.png

🖼️000001.png

🖼️000002.png

...

📁 instance # instance masks of each frame

    ️🖼️000000_mask.png

🖼️000001_mask.png

🖼️000002_mask.png

...

📁 nocs # nocs maps of each frame

🖼️000000.png 

🖼️000001.png

🖼️000002.png

... 

📁 normals # mesh normals maps of each frame

🖼️000000.png 

🖼️000001.png

🖼️000002.png

...

📁 pcd # point clouds of each frame

🖼️000000.npy

🖼️000001.npy

🖼️000002.npy

...

📁 rgb # RGB image of each frame

🖼️000000.png

🖼️000001.png

🖼️000002.png

...

📄 frame_metadata.json # extra information about each frame, eg.: object name, ID, 6D pose, scale

The main information about the point pairs annotations is contained in the PairAnnotations directory. Each file is named following the structure "PairID-SourceID-TargetID:category.json"

For questions and more information, send an email to sebeto@acin.tuwien.ac.at