---
license: apache-2.0
task_categories:
  - depth-estimation
language:
  - en
tags:
  - "360"
  - panorama
  - equirectangular
  - depth
  - da360
  - benchmark
  - geometry
pretty_name: DA360 Benchmark Test Sets
size_categories:
  - 1K<n<10K
arxiv: 2512.22819
configs:
  - config_name: metropolis
    description: Metropolis outdoor ERP test benchmark.
    data_files:
      - split: test
        path: archives/Metropolis.zip
  - config_name: matterport3d
    description: Matterport3D indoor ERP test benchmark.
    data_files:
      - split: test
        path: archives/Matterport3D.zip
  - config_name: stanford2d3d
    description: Stanford2D3D indoor ERP test benchmark (Area 5).
    data_files:
      - split: test
        path: archives/Stanford2D3D.zip
---

# DA360 benchmark test sets

Processed **equirectangular (ERP) RGB–depth** test splits used to evaluate [DA360: Depth Anything in 360°](https://antigravity-tech.github.io/DA360/) ([arXiv:2512.22819](https://arxiv.org/abs/2512.22819) · [code](https://github.com/antigravity-tech/DA360) · [demo](https://huggingface.co/spaces/hljiang/DA360)).

All subsets are **test-only** (no train/validation). Each benchmark is shipped as one zip under [`archives/`](archives/).

| Dataset | Source | Publication | Num images | Storage size | Note |
| ------- | ------ | ----------- | ---------- | ------------ | ---- |
| **Metropolis** | [Mapillary Metropolis](https://www.mapillary.com/dataset/metropolis) | [DA360 Sec. 3.3](https://arxiv.org/abs/2512.22819) | 3,000 | 3.1 GB | **New outdoor ERP benchmark.** Curated from about 10k synchronized Mapillary samples; quality-filtered; bottom ERP inpainted (UniFuse-style); cubemap depth reprojected to ERP. 16-bit depth, meters = pixel / 512. |
| **Matterport3D** | [Matterport3D](https://niessner.github.io/Matterport/) | [1] | 2,014 | 2.8 GB | Standard **18-scene indoor test split** (same protocol as prior panoramic works). `pano_skybox_color/` + `pano_depth/`. 16-bit depth, meters = pixel / 4000. |
| **Stanford2D3D** | [Stanford 2D-3D-S](http://buildingparser.stanford.edu/dataset.html) | [2] | 373 | 0.9 GB | Standard **Area 5** test set (`area_5a`, `area_5b`). 16-bit depth, meters = pixel / 512. |

**Total (all zips):** 6.8 GB. **6,387** RGB–depth ERP test pairs across the three benchmarks.

Evaluation in the paper uses **518×1036** ERP resolution. See the official repo for metrics and `bash scripts/evaluate.sh`.

---

## Download

Install the Hub client if needed:

```bash
pip install huggingface_hub
```

Download the full repository (or only `archives/`):

```bash
hf download hljiang/DA360 --repo-type dataset --local-dir /PATH/TO/DOWNLOAD --local-dir-use-symlinks False
```

Extract each benchmark into a shared `data/` folder (layout expected by [antigravity-tech/DA360](https://github.com/antigravity-tech/DA360)):

```bash
cd /PATH/TO/DOWNLOAD/archives
for z in Metropolis Matterport3D Stanford2D3D; do
  unzip -q "${z}.zip" -d ../data
done
```

One-shot (download zips from the Hub and extract to `./data`):

```bash
bash -c "$(curl -fsSL https://huggingface.co/datasets/hljiang/DA360/raw/main/scripts/prepare_data.sh)"
```

---

## Layout after unzip

```
data/
├── Metropolis/rgb/*.jpg          +  Metropolis/depth/*.png
├── Matterport3D/{scene}/pano_skybox_color|pano_depth/
└── Stanford2D3D/area_5{a,b}/rgb_|depth/
```

---

## Reading depth maps

| Dataset | Format | Metric depth |
| ------- | ------ | ------------- |
| Metropolis | 16-bit PNG | `depth_m = pixel_value / 512` |
| Matterport3D | 16-bit PNG | `depth_m = pixel_value / 4000` |
| Stanford2D3D | 16-bit PNG | `depth_m = pixel_value / 512` |

Reference loaders: [`demo/datasets/`](https://github.com/antigravity-tech/DA360/tree/main/demo/datasets) in the DA360 repository.

---

## Reference

**DA360 (this release).** Jiang, H., Song, Z., Lou, Z., Xu, R., Tan, M. *Depth Anything in 360°: Towards Scale Invariance in the Wild.* arXiv:2512.22819, 2025.

[1] Chang, A., Dai, A., Funkhouser, T., et al. Matterport3D: Learning from RGB-D Data in Indoor Environments. *3DV*, 2017.

[2] Armeni, I., Sax, S., Zamir, A. R., Savarese, S. Joint 2D-3D-Semantic Data for Indoor Scene Understanding. *arXiv:1702.01105*, 2017.

**Metropolis raw data.** Mapillary Metropolis dataset (2024). Used under the curation pipeline described in DA360 supplementary material (Sec. S1).

---

## Citation

If you use these benchmarks, please cite the DA360 paper and the original dataset sources above.