---
license: cc0-1.0
tags:
- robotics
- transparent-objects
- synthetic
---

# Syn-TODD original release mirror

**Complete: all 13 original files verified.**

Original author: Haoping Xu, University of Toronto.

Source: https://borealisdata.ca/dataset.xhtml?persistentId=doi:10.5683/SP3/LQKTXE (version 2.0).

This unofficial mirror preserves the 10 original dataset archives and 3 accompanying MVTrans model checkpoints, unchanged. The dataset contains synthetic transparent-object scenes for RGB-D, stereo, and multi-view perception.

License: [CC0 1.0](https://creativecommons.org/publicdomain/zero/1.0/), as recorded by Borealis.

Keep all train split ZIP volumes together, and all validation split ZIP volumes together. Use a split-ZIP-capable extractor starting from the corresponding `.zip` file. Do not extract `.z01`–`.z06` independently. The test ZIP is standalone.

Original implementation and paper: https://github.com/ac-rad/MVTrans and https://arxiv.org/abs/2302.11683 .

`manifest.json` records original MD5 checksums, file sizes and verified SHA-256 hashes.