Buy the pack — $149sheep.
- detr-r50-fp32 — calibrated
- F1 0.782
- same checkpoint — stock
- F1 0.708
delta +0.074 F1 — COCO val2017 sheep subset (65 images, 354 boxes), 800 px letterbox, 1-core x86, run 2026-09-30.
Sheep detection — livestock counting from a drone or a fence-post camera, the class a COCO model can actually see. We raced every Apache-2.0 architecture on a held-out flock set, calibrated the winner, and ship it as a pack with the numbers attached. The full table is below, and you can check our work on the real frame below.
The run, on a real frame.
This is the winning detector, unedited output: eleven sheep marked at the shipped operating point, every box on an animal, nothing invented. The image was not uploaded anywhere — the boxes below were drawn from the actual run and shipped with the page.
The race.
F1 at IoU 0.50 on the held-out set, stock threshold 0.5 flat versus calibrated. Latency is the median of a single CPU core.
| config | precision | recall | F1 stock | F1 calibrated | latency |
|---|---|---|---|---|---|
| detr-r50-fp32 — winner | 0.764 | 0.801 | 0.708 | 0.782 | 2126 ms |
| detr-r50-int8 | 0.747 | 0.712 | 0.663 | 0.729 | 1851 ms |
| rfdetr-n-fp32 | 0.783 | 0.629 | 0.644 | 0.697 | 422 ms |
| rfdetr-n-int8 | 0.778 | 0.614 | 0.624 | 0.686 | 324 ms |
run COCO val2017 sheep subset (65 images, 354 boxes), 800 px letterbox, 1-core x86, run 2026-09-30. the winning config's calibrated thresholds: sheep 0.80 — keyed by COCO class id as 20:0.80; NMS IoU 0.70. the values that produced the delta are in the pack.
In the pack.
- fp32 and INT8 ONNX weightsthe winning checkpoint, both precisions, exported and decode-tested
- verified decoderPython and JavaScript inference with the output contract checked against the reference implementation
- calibration reportinput resolution, per-class thresholds, NMS IoU and the measured delta against stock on COCO val2017 sheep subset (65 images, 354 boxes)
- Apache-2.0 attribution noticesno AGPL anywhere in the tree
- 10,000 hosted API callstop-ups never expire