LibreDFINEn

D-FINE-n (nano) detection weights, repackaged for LibreYOLO.

Reported COCO val2017 mAP50-95: 42.8.

Source

Derived from Peterande/D-FINE at the v1.0 release (dfine_n_coco.pth). This checkpoint is trained directly on the COCO dataset (no Objects365 pretraining is published for this size). Copyright (c) 2024 The D-FINE Authors. Licensed under the Apache License, Version 2.0.

The HGNetV2 backbone is copied from PaddlePaddle/PaddleClas. Copyright (c) 2024 PaddlePaddle Authors. Licensed under the Apache License, Version 2.0.

Modifications

Checkpoint metadata wrapping only — learned parameters are unchanged. The wrapper adds model_family, size, nc, and names fields so the unified LibreYOLO() factory routes correctly without filename heuristics. See weights/convert_dfine_weights.py in the LibreYOLO source repository.

Usage

pip install libreyolo

from libreyolo import LibreYOLO

model = LibreYOLO("LibreDFINEn.pt")
results = model("image.jpg", save=True)

License

Apache License 2.0. See the LICENSE and NOTICE files in this repository.