docs/en/datasets/depth/depth8.md
The Ultralytics Depth8 dataset is a compact monocular depth estimation dataset with 8 images sampled from the SUN RGB-D dataset: 4 for training and 4 for validation, drawn from its Kinect v1 and Kinect v2 captures (two per sensor in each split) for dense, artifact-free ground-truth depth maps. It is designed for rapid testing, debugging, and experimentation with YOLO26 depth estimation models and training pipelines — the 1.3 MB archive auto-downloads on first use, so yolo depth train data=depth8.yaml starts training within seconds.
!!! note
Depth8 is for pipeline testing only, not benchmarking — its 8 images are far too few for meaningful depth metrics. Use the full [NYU Depth V2](nyu-depth-v2.md) or [SUN RGB-D](sunrgbd.md) validation sets for representative results.
Depth8 follows the standard Ultralytics depth dataset layout: RGB images with paired .npy float32 depth maps in meters, matched by file stem.
depth8/
├── images/
│ ├── train/ # 4 images
│ └── val/ # 4 images
└── depth/
├── train/ # 4 float32 .npy depth maps
└── val/ # 4 float32 .npy depth maps
Depth values are real indoor sensor captures ranging from roughly 0.5 m to 4 m, well within the ≤10 m range of the full SUN RGB-D dataset.
The Depth8 dataset configuration is defined in a dataset YAML file, which specifies dataset paths, class names, and the download URL for the small packaged subset.
!!! example "ultralytics/cfg/datasets/depth8.yaml"
```yaml
--8<-- "ultralytics/cfg/datasets/depth8.yaml"
```
To train a YOLO26n-depth model on the Depth8 dataset with an image size of 640, use the following examples. For a full list of training options, see the YOLO Training documentation.
!!! example "Train Example"
=== "Python"
```python
from ultralytics import YOLO
# Load a pretrained YOLO26n-depth model
model = YOLO("yolo26n-depth.pt")
# Train the model on Depth8
results = model.train(data="depth8.yaml", epochs=100, imgsz=640)
```
=== "CLI"
```bash
# Train YOLO26n-depth on Depth8 using the command line
yolo depth train data=depth8.yaml model=yolo26n-depth.pt epochs=100 imgsz=640
```
Depth8 is sampled from SUN RGB-D — see the full SUN RGB-D dataset page for license details.
If you use the SUN RGB-D dataset in your research or development work, please cite the following paper:
!!! quote ""
=== "BibTeX"
```bibtex
@inproceedings{song2015sunrgbd,
title={SUN RGB-D: A RGB-D Scene Understanding Benchmark Suite},
author={Song, Shuran and Lichtenberg, Samuel P. and Xiao, Jianxiong},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2015}
}
```
The Ultralytics Depth8 dataset is designed for rapid testing and debugging of monocular depth estimation pipelines. With only 8 images (4 train, 4 val) in a 1.3 MB auto-downloading archive, it verifies the full train / val / predict cycle — paired depth-map loading, augmentation, loss computation, and metrics — in seconds, before scaling to a full dataset such as SUN RGB-D or NYU Depth V2.
Depth8 samples 8 images from SUN RGB-D's 9,245-train/1,090-val split, favoring Kinect v1/v2 captures with clean, dense depth maps. It uses the identical directory layout and .npy depth format, so a pipeline that runs on Depth8 runs unmodified on the full dataset — just point data= at depth-sunrgbd.yaml instead of depth8.yaml. Unlike the full dataset, Depth8 downloads in seconds and needs no conversion step.
No. Depth8 is too small for meaningful model comparison and is intended for training and evaluation pipeline checks. Use the full NYU Depth V2 or SUN RGB-D validation sets when you need representative depth estimation metrics.