Link to this sectionNYU Depth V2 Depth Dataset#

NYU Depth V2 is the standard indoor benchmark for monocular depth estimation. It consists of RGB-D video sequences of a wide variety of indoor scenes recorded with a Microsoft Kinect v1. It is the primary benchmark used to report YOLO26-Depth accuracy.

Link to this sectionKey Features#

• Captured with a Microsoft Kinect v1 RGB-D sensor.
• Covers a wide variety of real indoor scenes (homes, offices, classrooms, and similar spaces).
• Depth range up to approximately 10 m, typical of consumer indoor RGB-D capture.
• Evaluation is performed on the standard Eigen test split of 654 images.
• The primary benchmark for reporting monocular depth estimation accuracy.

Link to this sectionRole in YOLO26-Depth#

NYU Depth V2 is the primary zero-shot evaluation benchmark for the YOLO26-Depth family, and the headline metrics on the depth task page are reported on it. The published YOLO26-Depth models are not trained on NYU; although the dataset includes a train split, it is left unused and only held-out test results are reported.

Evaluation uses multi-scale and horizontal-flip test-time augmentation (TTA), followed by log-least-squares scale alignment between the predicted and ground-truth depth maps before metrics are computed.

Link to this sectionResults#

The table below reports the delta1 accuracy (percentage of pixels within a 1.25× threshold, higher is better) on the NYU Depth V2 Eigen test split by model size.

Link to this sectionDataset YAML#

A YAML (Yet Another Markup Language) file is used to define the dataset configuration. It contains information about the dataset's paths, classes, and other relevant information.

# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license

# NYU Depth V2 dataset for monocular depth estimation
# Documentation: https://cs.nyu.edu/~silberman/datasets/nyu_depth_v2.html
# 795 train + 654 val (Eigen test split) images, 480x640, indoor scenes, depth in meters
# Example usage: yolo depth train data=nyu-depth.yaml model=yolo26n-depth.pt
# parent
# ├── ultralytics
# └── datasets
#     └── nyu-depth  ← downloads here (≈1.5 GB)

# Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]
path: nyu-depth # dataset root dir (relative to Ultralytics settings 'datasets_dir')
train: images/train # train images (relative to 'path') 795 images
val: images/val # val images (relative to 'path') 654 images

# Depth maps are paired .npy files (float32, meters) under depth/<split>/, resolved by
# swapping '/images/' -> '/depth/' on each image path.

# Classes
nc: 1
names:
  0: depth

channels: 3

# Download script/URL (optional)
download: https://github.com/ultralytics/assets/releases/download/v0.0.0/nyu-depth.zip

Link to this sectionUsage#

To evaluate a YOLO26-Depth model on the NYU Depth V2 benchmark, you can use the following code snippets. For a comprehensive list of available arguments, refer to the model Validation page.

from ultralytics import YOLO

# Load a model
model = YOLO("yolo26x-depth.pt")  # load a pretrained depth model

# Evaluate on the NYU Depth V2 benchmark
results = model.val(data="nyu-depth.yaml")

Link to this sectionPretrained Models#

The YOLO26 depth family is evaluated zero-shot on the NYU Depth V2 benchmark. These models auto-download from the latest Ultralytics release, for example YOLO26x-depth from v8.4.0, and span a range of sizes (yolo26n/s/m/l/x-depth) for different accuracy and resource requirements.

Link to this sectionCitations and Acknowledgments#

If you use the NYU Depth V2 dataset in your research or development work, please cite the following paper:

@inproceedings{silberman2012indoor,
      title={Indoor Segmentation and Support Inference from RGBD Images},
      author={Silberman, Nathan and Hoiem, Derek and Kohli, Pushmeet and Fergus, Rob},
      booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
      year={2012}
}

We would like to acknowledge the authors for creating and maintaining this valuable resource for the computer vision community.