Streaming3D Dataset

This dataset contains assets used by the Streaming3D benchmark.

Overview

Subset	Files	Description
DL3DV	13,724	16 indoor/outdoor scenes from DL3DV
GSO30	20,099	30-object subset from Google Scanned Objects
NAVI	28,995	Navigation evaluation dataset
GSO	67,982	Full Google Scanned Objects dataset (1032 objects)

Download

You can download individual subsets using hf download:

hf download WalkerCH/Streaming3D DL3DV/
hf download WalkerCH/Streaming3D GSO30/
hf download WalkerCH/Streaming3D NAVI/
hf download WalkerCH/Streaming3D GSO/

Only the subset you request will be downloaded; you do not need to download the entire repository.

---

DL3DV

16 indoor/outdoor scenes from the DL3DV dataset. Each scene directory contains:

DL3DV/
  scene_<id>_<hash>_<name>/
    colmap/           # COLMAP reconstruction output (features, matches, sparse model)
    da3/              # DA3 format data
    images/           # Source images (PNG)
    sam3_masks/       # SAM3 segmentation masks
    split.json        # Train/test split
    transforms.json   # Camera transforms
  scenes.csv          # Scene metadata
  scenes selected.csv # Selected scene list

---

GSO30

GSO30 is a 30-object subset derived from Google Scanned Objects. Each object directory contains training renders, evaluation assets, and the original object mesh/material files.

Object List

alarm backpack bell blocks chicken cream elephant grandfather grandmother hat
leather lion lunch_bag mario oil school_bus1 school_bus2 shoe shoe1 shoe2
shoe3 soap sofa sorter sorting_board stucking_cups teapot toaster train turtle

Directory Structure

GSO30/
  <object_id>/
    meshes/
      model.glb
      model.obj
      model.mtl
      texture.png
    render_spiral_100/
      images/
        000.png ... 099.png
      masks/
        000.png ... 099.png
      model/
        000.png ... 099.png
        000.npy ... 099.npy
      transforms.json
      model_norm.obj
      model_norm.mtl
    render_mvs_25/
      model_norm.glb
      model_norm.obj
      model_norm.mtl
      model/
        000.png ... 024.png
        000.npy ... 024.npy

Some object folders also include auxiliary metadata, thumbnails, or legacy render folders. The benchmark protocol uses the paths above.

Usage

For training or reconstruction input, use all 100 images from:

GSO30/<object_id>/render_spiral_100/images/{000..099}.png

The corresponding masks are stored in:

GSO30/<object_id>/render_spiral_100/masks/{000..099}.png

Camera metadata for the 100 spiral views is available in:

GSO30/<object_id>/render_spiral_100/transforms.json
GSO30/<object_id>/render_spiral_100/model/{000..099}.npy

For evaluation, use the normalized GLB mesh and the 25 provided camera views from render_mvs_25:

GSO30/<object_id>/render_mvs_25/model_norm.glb
GSO30/<object_id>/render_mvs_25/model/{000..024}.npy

The matching reference renders for those views are:

GSO30/<object_id>/render_mvs_25/model/{000..024}.png

In short, the default protocol is:

1. Train or reconstruct from all render_spiral_100/images frames.
2. Evaluate by rendering or comparing against render_mvs_25/model_norm.glb using the 25 camera poses in render_mvs_25/model/*.npy.

---

GSO

Full Google Scanned Objects dataset containing 1032 objects.

GSO/
  <object_name>/
    images/           # Object renders (PNG)
    ...

---

NAVI

Navigation evaluation dataset with two difficulty levels.

NAVI/
  hard/              # Hard difficulty navigation scenes
  normal/            # Normal difficulty navigation scenes
  src_code/          # Source code for evaluation
  README.md          # NAVI-specific documentation