RoboTwin2.0 Data Upscaled
This dataset is derived from the pre-collected RoboTwin2.0 data released by
https://huggingface.co/datasets/TianxingChen/RoboTwin2.0.
We processed the original RoboTwin2.0 training data with two additional visual modalities:
RGB upscaling: head-camera RGB frames were upscaled using FlashVSR.
Verified resolution: 320x240 -> 640x480
Depth estimation: monocular depth maps were extracted from the head-camera RGB frames using Depth-Anything-V2.
Stored as float16 NumPy… See the full description on the dataset page:
https://huggingface.co/datasets/Soonwoo/RoboTwin2.0-data-upscaled.
