Skip to content

Record a sky camera around the clock, keep the footage on your own disk, and find what flew past.

3.3 KB
Raw Blame
{
  "family": "openskycam-night",
  "version": "yolo26s-1920-2026-10-25",
  "name": "openskycam night · YOLO26s · 1920",
  "description": "Night detector for planes and meteors seen by a fixed sky camera, run on the whole frame at 1920 px. It descends from Ultralytics YOLO26s: an earlier openskycam detector was fine-tuned on 4,036 reviewed night frames from one 4K camera. Frames a reviewer accepted keep their boxes. Rejected frames whose box stayed in one place (stars, glare, building and tower lights) train as empty sky, at most 8 per night and spot in the sky, so the model learns not to flag them. Meteor frames are repeated 3 times in training. Train, validation and test splits are whole nights.",
  "task": "object-detection",
  "adapter": "ultralytics",
  "classes": [
    {
      "id": 0,
      "name": "Plane"
    },
    {
      "id": 1,
      "name": "Meteorite"
    }
  ],
  "input": {
    "channels": 3,
    "color_space": "RGB",
    "kind": "image",
    "preprocessing": "cv2.INTER_AREA to the long side, as the Workshop runner does, then Ultralytics letterbox",
    "training_size": 1920
  },
  "evaluations": [
    {
      "label": "Validation during training",
      "split": "validation",
      "input_size": 1920,
      "dataset": "792 images from 2 held-out nights (686 planes, 69 meteors)",
      "metrics": {
        "epoch": 23,
        "metrics/mAP50(B)": 0.94697,
        "metrics/mAP50-95(B)": 0.65347,
        "metrics/precision(B)": 0.92388,
        "metrics/recall(B)": 0.92624
      },
      "independently_reproduced": false
    },
    {
      "label": "Held-out test nights, confidence 0.3",
      "split": "test",
      "input_size": 1920,
      "dataset": "1,245 images from 7 held-out nights (891 planes, 157 meteors)",
      "metrics": {
        "plane recall": 0.955,
        "meteorite recall": 0.904,
        "stationary lights flagged": 0.074,
        "extra detections per labelled image": 0.049,
        "previous model": {
          "plane recall": 0.925,
          "meteorite recall": 0.866,
          "stationary lights flagged": 0.485,
          "extra detections per labelled image": 0.031
        }
      },
      "independently_reproduced": false
    }
  ],
  "caveats": [
    "Trained on one camera at one site. Other cameras, lenses, exposure settings and skies will see lower recall and more false alarms; review its detections on your own footage before relying on it.",
    "Night only: every training, validation and test frame was taken between 22:00 and 05:00 local time. It has not seen daylight or twilight.",
    "The test labels are frames a person reviewed among the previous model's detections, so test recall measures keeping what that model found, not finding objects both models miss.",
    "A detection counts as a match when it overlaps a label (IoU 0.1) or its centre is within one box length of the label's.",
    "Rejected frames of moving objects, satellites among them, were left out of training so the model is not taught to ignore movement. It has no satellite class and can label a satellite as a Plane.",
    "Trained with Ultralytics 8.4.165. It loads and gives the same detections with Ultralytics 8.4.58, the version the processor installs.",
    "The metrics come from the training notebook, not from a Workshop run."
  ],
  "dependencies": [],
  "created_at": "2026-10-25T21:04:01+00:00"
}