{"success":true,"data":{"author":"Antoine Carreaud, Shanci Li, Malo De Lacour, Digre Frinde, Jan Skaloud, Adrien Gressin (HEIG-VD)","bucketAccessType":"public","bucketName":"heig-vd-geo/GridNet-HD","bucketPrefix":"","category":"Earth Observation & Geospatial","cloudProvider":"huggingface","createdAt":"2026-11-21T20:45:58.871Z","description":"GridNet-HD (Grid Network at High Density and High Resolution) is a multimodal dataset of co-registered drone (UAV) LiDAR point clouds and high resolution RGB photographs of electrical power line corridors, built by the geomatics group of HEIG-VD (School of Engineering and Management Vaud, Switzerland) for 3D semantic segmentation with LiDAR-image fusion. It covers 36 geographic zones, one folder each (t1z4, t1z5a and so on), holding images/ (5280 by 3956 pixel JPGs), masks/ (single-channel PNG semantic masks, one per image), lidar/ (one dense .las point cloud per zone with a ground_truth class field) and pose/ (camera positions, orientations and intrinsics in Agisoft Metashape convention) so points can be projected into images. Labels are grouped into 12 semantic classes: pylon, conductor cable, structural cable, insulator, high, low and herbaceous vegetation, rock, gravel and soil, road, water, building, and unlabeled. split.json gives the official train and test zones (about 70 and 30 percent of points); test labels are withheld and scored by mIoU on a Hugging Face leaderboard. About 13.5 GB. License CC BY 4.0. Cite Carreaud, Li, De Lacour, Frinde, Skaloud and Gressin (2026), arXiv 2601.13052.","displayTitle":"GridNet-HD: UAV LiDAR and Image Fusion Dataset for 3D Semantic Segmentation of Power Line Infrastructure (HEIG-VD)","huggingFaceUrl":"https://huggingface.co/datasets/heig-vd-geo/GridNet-HD","id":"e092ed02-282c-5563-9337-aeffb51f1b63","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/heig-vd-geo/GridNet-HD","objectCountEstimate":13053,"probeStatus":"ok","referenceUrl":"https://huggingface.co/datasets/heig-vd-geo/GridNet-HD","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"heig-vd-geo/GridNet-HD\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"heig-vd-geo/GridNet-HD\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=heig-vd-geo, license:cc-by-4.0, size_categories:10K<n<100K, format:imagefolder, downloads=11210"],"format":"mixed","generatedAt":"2026-11-21T20:48:48.382Z","layout":{"example":"figures/3d_segmented.png","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"89504e470d0a1a0a0000000d49484452000002e7000003790806000000ab1c954f0000000473424954080808087c0864880000001974455874536f66747761726500676e6f6d652d73637265656e73686f74ef03bf3e0000002e744558744372656174696f6e2054696d6500467269203136204d617920323032352030383a30363a323620414d2043455354b091eba90000200049444154789cecbd3da825d7b9f7b97411172698648c4034ba120e6478b94ce03e418f52333037f17be50f8c1dc8c905071714bc88e6e00e142890691a3181c081c189155818d99647c91b39d5dbc169078379611c0c2d8b4620fc2613182ec63d41f7dafdece7ac8f677d55adaafafda0e973f6ae5a557bd7de75fef5affff3ace71e3f7efcd80100000000c0eafcc3da3b0000000000004f409c03000000004c02e21c00000000601210e7000000000093803807000000009804c43900000000c02420ce01000000002601710e0000000030098873000000008049409c03000000004c02e21c00000000601210e7000000000093803807000000009804c43900000000c02420ce01000000002601710e0000000030098873000000008049409c03000000004c02e21c00000000601210e7000000000093803807000000009804c43900000000c02420ce01000000002601710e0000000030098873000000008049409c034cc8dffef4cdb3","encoding":"hex","key":"figures/3d_segmented.png","truncatedAt":512}],"sampleObjects":["figures/3d_segmented.png","figures/class_distribution.png","figures/corresponding_image.png","figures/raw_3d.png","split.json"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\nlanguage:\n- en\ntags:\n- 3d\n- Image\npretty_name: GridNet-HD\nsize_categories:\n- 1B<n<10B\ntask_categories:\n- feature-extraction\n\n---\n\n# 🗂 GridNet-HD dataset\n\n## 1. Introduction\n\nThis dataset was developed for **3D semantic segmentation task** using both **images and 3D point clouds** specialized on electrical infrastructure.\n**Grid** (electrical) **Net**work at **H**igh **D**ensity and High Resolution represents the first Image+LiDAR dataset accurately co-referenced in the electrical infrastructure domain.\nThis dataset is associated with a public leaderboard hosted on Hugging Face Spaces, available at: [leaderboard](https://huggingface.co/spaces/heig-vd-geo/GridNet-HD-Leaderboard).\n\nThe dataset is associated with the following paper:\n\n```\n@misc{gridnet-hd-dataset,\n      title={GridNet-HD: A High-Resolution Multi-Modal Dataset for LiDAR-Image Fusion on Power Line Infrastructure}, \n      author={Antoine Carreaud and Shanci Li and Malo De Lacour and Digre Frinde and Jan Skaloud and Adrien Gressin},\n      year={2026},\n      eprint={2601.13052},\n      url={https://arxiv.org/abs/2601.13052}, \n}\n```\n\nThis repository hosts the official data splits and resources used in the experiments reported in the paper.\n\n| Raw 3D | 3D segmented | Corresponding image |\n| --- | --- | --- |\n| ![Raw 3D point cloud](figures/raw_3d.png) | ![Segmented 3D point cloud](figures/3d_segmented.png) | ![Corresponding RGB image](figures/corresponding_image.png) |\n\n---\n\n## 2. Dataset Structure\n\nThis dataset consists of 36 geographic zones, each represented by a folder named after its area code (e.g. t1z4, t1z5a, etc.).\n\nEach zone contains aligned multimodal data (images, segmentation masks, LiDAR point cloud, and camera parameters), enabling high-precision image-to-3D projection for multimodal fusion 3D semantic segmentation task.\n\nA split.json file at the root of the dataset defines the official train/test partition of the zones.\n\nTo ensure fair evaluation on the **official test set**, ground truth annotations are not provided for either the images or the LiDAR point clouds.\nInstead, participants must submit their predictions to the [leaderboard](https://huggingface.co/spaces/heig-vd-geo/GridNet-HD-Leaderboard), where the official metrics (mIoU) are automatically computed against the hidden labels.\n\n### 📁 Folder layout\n```\ndataset-root/\n├── t1z5b/\n│   ├── images/           # RGB images (.JPG)\n│   ├── masks/            # Semantic segmentation masks (.png, single-channel label)\n│   ├── lidar/            # LiDAR point cloud (.las format with field \"ground_truth\")\n│   └── pose/             # Camera poses and intrinsics (text files)\n├── t1z6a/\n│   ├── ...\n├── ...\n├── split.json            # JSON file specifying the train/test split\n└── README.md\n```\n### 🧾 Contents per zone\n\nInside each zone folder, you will find:\n- 📷 images/\n\t- High-resolution RGB images (5280x3956) (.JPG)\n\t- Captured from a UAV\n\n- 🏷️ masks/\n\t- One .png mask per image, same filename as image\n\t- Label-encoded masks (1 channel)\n\n- 🌍 lidar/\n\t- Single .las file for the entire zone captured from a UAV\n\t- Contains 3D point cloud data at high denisty with semantic ground_truth labels (stored in field named \"ground_truth\")\n\n- 📌 pose/\n\t- camera_pose.txt: Camera positions and orientations per image (using Metashape Agisoft convention, more details in paper)\n\t- camera_calibration.xml: Camera calibration parameters (using Metashape Agisoft calibration model)\n\n---\n\n## 3. Class Grouping\n\nOriginal classes have been grouped into **12 semantic groups** as follows:\n\n| Group ID | Original Classes  | Description                       |\n|:--------:|:-----------------:|:---------------------------------:|\n| 0        | 0,1,2,3,4         |   Pylon                           |\n| 1        | 5                 |   Conductor cable                 |\n| 2        | 6,7               |   Structural cable                |\n| 3        | 8,9,10,11         |   Insulator                       |\n| 4        | 14               ","encoding":"utf8","key":"README.md","name":"README.md","size":13457,"truncated":true}],"sourceObjectKey":"figures/3d_segmented.png","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"heig-vd-geo/GridNet-HD\", filename=\"figures/3d_segmented.png\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"figures/3d_segmented.png","totalObjects":13053},"summary":"GridNet-HD (Grid Network at High Density and High Resolution) is a multimodal dataset of co-registered drone (UAV) LiDAR point clouds and high resolution RGB photographs of…","title":"GridNet-HD: UAV LiDAR and Image Fusion Dataset for 3D Semantic Segmentation of Power Line Infrastructure (HEIG-VD)","updatedAt":"2026-11-21T21:54:08.305Z","visibility":"public","websiteUrl":"https://arxiv.org/abs/2601.13052"}}