{
  "annotation_id": "6c7d7ad1-2020-440f-93ec-9fc05ddc3197",
  "date_created": "2026-02-17T05:53:24.056000Z",
  "date_modified": "2026-02-17T05:53:24.056000Z",
  "file_hash": "948a2dd330b894887b588e9593ff27e07bbb80e602fa4e8fd4e82a34a65883ad",
  "private": false,
  "record": {
    "abstract": "Online high-definition (HD) map construction is an essential part of a safe and robust end-to-end autonomous driving (AD) pipeline. Onboard camera-based approaches suffer from limited depth perception and degraded accuracy due to occlusion. In this work, we propose SatMap, an online vectorized HD map estimation method that integrates satellite maps with multi-view camera observations and directly predicts a vectorized HD map for downstream prediction and planning modules. Our method leverages lane-level semantics and texture from satellite imagery captured from a Bird\u0027s Eye View (BEV) perspective as a global prior, effectively mitigating depth ambiguity and occlusion. In our experiments on the nuScenes dataset, SatMap achieves 34.8% mAP performance improvement over the camera-only baseline and 8.5% mAP improvement over the camera-LiDAR fusion baseline. Moreover, we evaluate our model in long-range and adverse weather conditions to demonstrate the advantages of using a satellite prior map. Source code will be available at https://iv.ee.hm.edu/satmap/.",
    "arxiv_id": "2601.10512",
    "authors": [
      "Kanak Mazumder",
      "Fabian B. Flohr"
    ],
    "categories": [
      "cs.CV",
      "cs.AI",
      "cs.LG"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "title": "SatMap: Revisiting Satellite Maps as Prior for Online HD Map Construction",
    "url": "https://arxiv.org/abs/2601.10512",
    "version": "v2"
  },
  "schema_id": "dorsal/arxiv",
  "source": {
    "execution_id": "d0781e60-3d60-4c24-aab5-43d192399af4",
    "id": "arXiv Dataset",
    "type": "Model",
    "variant": "snapshot-2026-01-17",
    "version": "0.1.0"
  },
  "user_id": 1000002
}