Version 1.0

Highlights

• New datasets added with stronger pre-trained checkpoints for better zero-shot capabilities
• Single Inference demo script for cross-modal instance retrieval
• Move from using labelmap color space as the features for scene pointclouds to captured raw RGB colors
• Move to more efficient storage format for preprocessed data

What's Changed?

• Release of MultiScan+ARKitScenes code, by @gauravpradeep, resolves #1 , #3 and closes #2 & #4
• Support for scene alignment and normalization across datasets, by @gauravpradeep, closes #10
• Demo for object level single inference for in-the-wild cross-modal retrieval, by @gauravpradeep, resolves #13

This update ensures that CrossOver is now compatible with four major datasets, and increases ease of use of our models.