Download the dataset

The full Tree Vision benchmark is hosted on Hugging Face: all modalities, the calibration JSON for every capture set and the merged point clouds.

Download from Hugging Face → Look at the examples first

Or copy the link: https://huggingface.co/datasets/BennyLin01/real_world_virtual_tree

What you get

Modality	Description
Left RGB `left`	Left-camera RGB renderings of the stereo pair.
Right RGB `right`	Right-camera RGB renderings forming the stereo pair (63 mm baseline).
Left Depth `left_depth`	Per-pixel metric depth. EXR/NPY and a 16-bit millimetre PNG for ground truth, *_vis.png for preview.
Right Depth `right_depth`	Per-pixel metric depth for the right camera.
Left Disparity `left_disparity`	Disparity in pixels (left frame), 16-bit PNG and float NPY.
Right Disparity `right_disparity`	Disparity in pixels (right frame).
Left Surface Normals `left_normal`	World-space surface normals. EXR/NPY for ground-truth, *_vis.png for preview.
Right Surface Normals `right_normal`	World-space surface normals for the right view.
Left Instance Segmentation `left_seg`	Per-instance segmentation: colourised PNG, integer ID PNG, target mask and overlay.
Right Instance Segmentation `right_seg`	Per-instance segmentation for the right view.
Left Semantic Classes `left_class`	Semantic class map: _class.png holds the integer class per pixel, _class_color.png is the colourised preview.
Right Semantic Classes `right_class`	Semantic class map for the right view.
Left Detection Labels `left_labels`	YOLO annotations for the target tree: normalised polygons in _labels.txt, boxes in _labels_bbox.txt.
Right Detection Labels `right_labels`	YOLO polygons and boxes for the right view.
Point Clouds `pointcloud`	Dense point clouds (.ply), per frame and/or merged across the orbit, in raw and coloured variants.

Total size grows linearly with the number of capture sets. The sample published on this site is 118 files across 2 capture set(s), 202.8 MB.

Directory layout

Every capture set unpacks to the same structure:

tree_NN/
├── camera_poses.json          calibration + orbit description
├── segmentation_info.json     target instance metadata
├── left/            right/            8-bit RGB PNG
├── left_depth/      right_depth/      EXR + NPY (metres) + 16-bit mm PNG
├── left_disparity/  right_disparity/  NPY (px) + 16-bit PNG
├── left_normal/     right_normal/     EXR + NPY (world space)
├── left_seg/        right_seg/        colour PNG + ID PNG (+ mask / overlay)
├── left_class/      right_class/      semantic class PNG + colourised preview
├── left_labels/     right_labels/     YOLO polygons + boxes (.txt)
└── pointcloud/                        per-frame and/or merged PLY

Not every release carries every folder: semantic classes and detection labels are new in the HD720 render, and the 1080P export ships float EXR and NPY where HD720 ships 16-bit PNG. A capture set's own page lists exactly what it contains.

Point DATASET_ROOT at the directory that holds these folders and this site will browse your local copy directly.

Licence

By downloading you agree to use the dataset for non-commercial, academic purposes under CC BY-NC-SA 4.0. Attribution is required — the BibTeX entry is in the citation section on the home page.

Publishing derived work, redistributing modified copies or asking about commercial use? Get in touch — we are happy to help.