TartanAir: a dataset to push the limits of visual SLAM
- Provider
- Carnegie Mellon University AirLab (Wenshan Wang et al.)
- License
- CC-BY-4.0 (source)
- Size
- 290 files
- Stored at
- Hugging Face (https://huggingface.co/datasets/theairlabcmu/tartanair)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Computer Vision
- Source
- https://arxiv.org/abs/2003.14338
About TartanAir: a dataset to push the limits of visual SLAM
TartanAir is Carnegie Mellon University AirLab's synthetic benchmark for visual SLAM and robot navigation (Wang et al., IROS 2020, arXiv:2003.14338). The data was collected in photo-realistic simulation environments with varied lighting, weather and moving objects, along camera trajectories designed to be hard for state-of-the-art visual odometry and SLAM, including challenging viewpoints and diverse motion patterns. Because it is simulated, every frame carries precise multi-modal ground truth: stereo RGB images, depth images, semantic segmentation, optical flow, camera poses and LiDAR point clouds. Measured 2026-11-14: 18 environment folders (abandonedfactory, abandonedfactory_night, amusement, carwelding, endofworld, gascola, hospital, japanesealley, neighborhood, ocean, office, office2, oldtown, seasidetown, seasonsforest, seasonsforest_winter, soulcity, westerndesert), each split into Easy and Hard trajectory sets stored as zip archives per modality (image_left, image_right, depth_left, depth_right, seg_left, seg_right, flow_flow, flow_mask); the Hub reports about 1.1 TB. Unzip and read PNG images and NumPy depth and flow arrays, or use the tartanair Python toolkit. Uses: stereo matching, monocular depth, optical flow, visual odometry and SLAM training and evaluation. The AirLab licenses the dataset under CC BY 4.0; cite the TartanAir paper.
How to load TartanAir: a dataset to push the limits of visual SLAM in Python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="theairlabcmu/tartanair", filename="abandonedfactory/Easy/depth_left.zip", repo_type="dataset")
print(path)Data structure
Files
Example file: abandonedfactory/Easy/depth_left.zip
Analyze this dataset without downloading it
On Scigantic, TartanAir: a dataset to push the limits of visual SLAM opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Computer Vision datasets
- Vipe Web 360
- SKZ 3D Printing Defect Detection, Nozzle Camera: YOLO-Labeled FFF Print Errors
- Nemotron Personas Korea
- The Metropolitan Museum of Art Open Access: collection metadata and images
- Quick, Draw! Dataset: 50 Million Hand-Drawn Sketches in 345 Categories as Stroke Vectors and 28x28 Bitmaps (Google Creative Lab)
- uCO3D: UnCommon Objects in 3D
- Textvqa
- YouTube UGC Dataset: User-Generated Content Video Clips with Mean Opinion Scores for Video Quality Assessment and Compression (Google)