About Vision Analysis

Credible and deep analysis of computer vision models.

Mission

Vision Analysis provides credible benchmarks for object detection models. All benchmarks are run using vision-analysis-benchmark on top of LibreYOLO.

Submission Flow

Community contributors run the harness locally, produce a submission JSON, and open a pull request adding that file under submissions/.

CI validates each submission against the schema and the current support matrix, then rebuilds the canonical benchmark dataset consumed by the website.

The site renders only from the generated canonical dataset. Raw submission files stay in the repo for review and provenance, but the public UI reads the rebuilt verified snapshot.

Methodology

All models are evaluated using the standard pycocotools protocol on coco-val2017-mini500, a frozen 500-image subset of COCO val2017 (80 classes).

We measure end-to-end latency including preprocessing, inference, and postprocessing when the runtime exposes those phases.

Every published result should include the benchmark config, actual input size, hardware identity, runtime/provider, and the supporting LibreYOLO commit so the number is reproducible.

The full protocol (warmup, timing, evaluation settings, environment capture, and known limitations) is documented on the Methodology page.

Run any model with one line

LibreYOLO has the best catalogue of state-of-the-art detectors, all behind one MIT-licensed Python API.

from libreyolo import LibreYOLO, SAMPLE_IMAGE

# LibreYOLO has the best catalogue of state-of-the-art models.
model = LibreYOLO("LibreRFDETRl.pt")           # RF-DETR-L (transformer flagship)
results = model(SAMPLE_IMAGE, save=True)        # run inference, save the annotated image

# Swap in any other model, same one-line API (weights auto-download):
#   LibreYOLO("LibreYOLO9c.pt")      # YOLO9-C
#   LibreYOLO("LibreYOLOXx.pt")      # YOLOX-X
#   LibreYOLO("LibreDFINEx.pt")      # D-FINE-X
#   LibreYOLO("LibreRTDETRr50.pt")   # RT-DETR-R50