{"success":true,"data":{"author":"AD Assurance Lab, Western Michigan University, with Mcity","bucketAccessType":"public","bucketName":"AD-Assurance-Lab/winter-driving-dataset","bucketPrefix":"","category":"Robotics","cloudProvider":"huggingface","createdAt":"2026-11-21T22:15:32.940Z","description":"Winter automated-driving data from the AD Assurance Lab at Western Michigan University with Mcity, in two modules. MI-Snow1000 is a lane detection benchmark for snow-covered roads: 4,888 dashcam frames (1280x720) from over 1,000 miles of winter driving in Michigan, Colorado and Wyoming, each labeled with the left and right ego-lane boundaries in TuSimple format, following tire tracks and plow lines where snow hides the paint; 535 frames are deliberately labeled with no lane. A sequence-disjoint split has 3,105 training, 887 validation and 337 test frames, with per-frame snow, glare and curvature tags, and SCNN, PolyLaneNet, LaneATT and CLRNet checkpoints are included in ONNX and PyTorch. MCity-WSPI (Winter Wonderset) pairs 25 Hz front-camera frames from a Ford Mustang Mach-E at Mcity in January and February 2026 with CAN steering, throttle, brake torque and wheel speeds, a 6-DoF IMU, RTK GPS and MARWIS road-friction and road-condition readings; raw per-topic CSV exports of 62 runs are under raw/. Images ship as per-run zips; about 164 GB (measured 2026-11-21). Annotations and loaders are on the lab's GitHub. Released under CC BY 4.0.","displayTitle":"AD Winter Driving Dataset: MI-Snow1000 Snow-Covered Lane Detection and MCity-WSPI Vehicle Dynamics on Snow and Ice (WMU AD Assurance Lab)","huggingFaceUrl":"https://huggingface.co/datasets/AD-Assurance-Lab/winter-driving-dataset","id":"020b0094-547f-58a6-bceb-83d2cdfb33b7","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/AD-Assurance-Lab/winter-driving-dataset","objectCountEstimate":989,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"AD-Assurance-Lab/winter-driving-dataset\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"AD-Assurance-Lab/winter-driving-dataset\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=AD-Assurance-Lab, license:cc-by-4.0, size_categories:100K<n<1M, downloads=1180"],"format":"mixed","generatedAt":"2026-11-21T22:31:27.107Z","layout":{"example":"raw/marwis_kalamazoo_2022-23/MARWIS 10722_20221220153413-20221220155329.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"date,time,latitude,longitude,altitude,course,speed,Dew Point [°C],Ambient Temperature [°C],Ambient Rel. Humidity [%],Surface Temperature [°C],Road Condition Lufft [logic],Friction [],Ice Percent [%],Water Film Height [µm],\r\n12/20/2022,3:34:13 PM,42.254695119,-85.638394838,287.24,299,-4,-0.26,14.97,34.35,10.99,1.00,0.81,0.00,43.50,\r\n12/20/2022,3:34:20 PM,42.254658273,-85.638389986,287.20,270,-4,-0.78,14.99,33.06,10.96,1.00,0.81,0.00,42.46,\r\n12/20/2022,3:34:29 PM,42.254671069,-85.638405449,287.22,279,-","encoding":"utf8","key":"raw/marwis_kalamazoo_2022-23/MARWIS 10722_20221220153413-20221220155329.csv","truncatedAt":512}],"sampleObjects":["images/mcity_wspi/feb23-straight-10_images.zip","images/mcity_wspi/feb23-straight-15_images.zip","images/mcity_wspi/feb23-straight-20_images.zip","images/mcity_wspi/feb23-straight-25_images.zip","images/mcity_wspi/feb23-straight-30_images.zip"],"sidecars":[{"bytes":"---\npretty_name: \"AD Winter Driving Dataset: MI-Snow1000 lane detection and MCity-WSPI physics\"\ntask_categories:\n- image-segmentation\n- object-detection\ntags:\n- mi-snow1000\n- lane-detection\n- snow\n- winter-driving\n- adverse-weather\n- tusimple\n- autonomous-driving\n- adas\n- vehicle-dynamics\n- weather-telemetry\nlicense: cc-by-4.0\nsize_categories:\n- 100K<n<1M\n---\n\n# AD Winter Driving Dataset: MI-Snow1000 Lane Detection and MCity-WSPI Physics\n\nThis is the official Hugging Face repository hosting the image binary packages and pre-trained benchmarks for the **AD Winter Driving Dataset**, created by the **AD Assurance Lab** in partnership with **Western Michigan University** and **MCity**. It has two modules:\n\n- **MI-Snow1000** is a lane detection dataset for snow-covered roads: 4,888 dashcam frames (1280x720) from over 1,000 miles of winter driving in Michigan, Colorado and Wyoming, each annotated with the left and right boundaries of the ego lane in TuSimple format. Where snow buries the paint, the boundaries follow the drivable corridor shown by tire tracks and plow lines. It includes a sequence-disjoint train, validation and test split (3,105, 887 and 337 frames), per-frame snow, glare and curvature tags, and trained SCNN, PolyLaneNet, LaneATT and CLRNet checkpoints in PyTorch and ONNX. License CC BY 4.0.\n- **MCity-WSPI** pairs front-camera frames with vehicle dynamics, IMU, GPS and road friction telemetry recorded on snow and ice at Mcity.\n\nFor the annotations, splits, synchronized CSV metadata, loaders, and developer documentation, please visit the main GitHub repository:\n👉 [GitHub: AD-Assurance-Lab/dataset--winter-driving--main](https://github.com/AD-Assurance-Lab/dataset--winter-driving--main)\n\n---\n\n## MI-Snow1000 in brief\n\n| | |\n|---|---|\n| Frames | 4,888 annotated dashcam frames, 1280x720 JPEG |\n| Labels | Left and right boundary of the ego lane, TuSimple format (56 rows, 160 to 710; -2 where a boundary is absent) |\n| Benchmark split | `metadata/sequence_disjoint/` on GitHub: 3,105 training, 887 validation, 337 test frames, no drive shared between partitions |\n| Condition tags | `metadata/condition_tags.json` on GitHub: snow severity, glare, curvature, single boundary |\n| Models | `models/` here: SCNN, PolyLaneNet, LaneATT, CLRNet at 320x180, trained on the benchmark training partition (seed 0) |\n| License | CC BY 4.0 |\n\n**Frames with no lane boundaries are intentional.** Where the annotators judged that no boundary could be placed with confidence, for example on a fully snow-covered road with no tracks, the frame was kept with no lanes labeled, so that a detector learns to output nothing rather than guess. These are 535 of the 4,888 frames (11%). Treat them as frames whose correct output is no lane, not as missing labels.\n\nTo ask for an image to be removed, for example one that shows your face or license plate, open an issue on the GitHub repository with the frame name.\n\n---\n\n## Dataset Structure\n\nThis repository is structured to separate tabular metadata, pre-trained models, and large image binaries:\n\n```text\nwinter-driving-dataset/\n├── metadata/                 # Annotations, splits and synchronized CSVs (on GitHub)\n├── models/                   # Pre-trained lane detection checkpoints\n│   ├── onnx/                 # ONNX float32/float16 format checkpoints\n│   └── pytorch/              # PyTorch .pt / .pth checkpoints\n└── images/\n    ├── mcity_wspi/           # Individual zip packages for each dynamics sequence\n    │   ├── jan27-downtown-1_images.zip\n    │   ├── feb23-straight-10_images.zip\n    │   └── ...\n    └── reva_perception/\n        └── non_labeled_set/  # Unlabeled MI-Snow1000 drive frames, one zip per session\n└── raw/\n    └── mcity_wspi/           # Raw recordings the synchronized CSVs were made from\n        ├── csv/<day>/        # Per-topic ROS message exports, 62 runs over four days\n        ├── jan20/            # Shakedown day: OxTS and vehicle CSVs, video, topic list\n        ├── bags/             ","encoding":"utf8","key":"README.md","name":"README.md","size":8016,"truncated":true}],"sourceObjectKey":"raw/marwis_kalamazoo_2022-23/MARWIS 10722_20221220153413-20221220155329.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/AD-Assurance-Lab/winter-driving-dataset/raw/marwis_kalamazoo_2022-23/MARWIS 10722_20221220153413-20221220155329.csv\")\ndf.head()","starterSamplePath":"images/mcity_wspi/feb23-straight-10_images.zip","totalObjects":989},"summary":"Winter automated-driving data from the AD Assurance Lab at Western Michigan University with Mcity, in two modules.","title":"AD Winter Driving Dataset: MI-Snow1000 Snow-Covered Lane Detection and MCity-WSPI Vehicle Dynamics on Snow and Ice (WMU AD Assurance Lab)","updatedAt":"2026-11-22T01:13:04.801Z","visibility":"public","websiteUrl":"https://github.com/AD-Assurance-Lab/dataset--winter-driving--main"}}