Auspex Seg

Auspex Seg predicts a polygon per object, together with the class and box that identify it. It trains from random initialisation: nothing is downloaded and no pretrained weights are required, so a model you train carries nothing of anyone else's.

Boxes come free

You annotate polygons; you do not annotate boxes. An annotation carrying only a segmentation is complete — the box is taken from the extent of its rings.

The box still exists inside the model, and that is deliberate. Mask prototypes are shared across the whole image, so the only thing that makes one of them an instance is its box. Every predicted polygon therefore arrives with a box and a score beside it, at no extra labelling cost.

Use

from auspex_seg import Auspex

model = Auspex(labelspace="labelspace.yaml") model.train(data="data.yaml", epochs=150) results = model.predict("images/", save=True) ```

Or the console command: auspex-seg train --data data.yaml --labelspace labelspace.yaml

Formats

Reads COCO JSON, CVAT-for-images 1.1 XML and Pascal VOC XML, mixed freely in one training run. Predictions export back to CVAT or COCO for pre-labelling, and tiled inference keeps small objects detectable on very large images.

Runs on CPU

Training and prediction both work on a CPU-only install — no NVIDIA driver required. img_size is the lever that matters there, and --no-refine cuts prediction time roughly fivefold by skipping per-instance mask refinement.

Trial

Training is free for 30 days in full. Prediction, validation, calibration and export are never gated — they keep working afterwards, forever, on any model you have already trained. Credentials verify offline, so training works on air-gapped machines.