LibreCenterNetresdcn18

CenterNet ResNet-18 with deformable upsampling COCO detector, repackaged for LibreYOLO.

from libreyolo import LibreYOLO

model = LibreYOLO("LibreCenterNetresdcn18.pt")
results = model.predict("image.jpg")

Source

Derived from xingyizhou/CenterNet at commit 4c50fd3a46bdf63dbf2082c5cbb3458d39579e6c. Copyright (c) 2019 Xingyi Zhou. The source implementation is MIT licensed.

Official checkpoint: ctdet_coco_resdcn18.pth
Official SHA-256: f9e413f91cdb235adbcb41c5c4052b8f7ff53999374048949789c29d6df18eaa
Published COCO test-dev AP without test-time augmentation: 28.1.

Modifications

The data-parallel module. prefix was removed and LibreYOLO v1 checkpoint metadata was added. Learned tensors are unchanged. The native graph strictly loads the official state dict and its hm, wh, and reg outputs are bit-exact against the pinned implementation. LibreYOLO replaces the legacy DCNv2 extension with torchvision deformable convolution.

See weights/convert_centernet_weights.py and docs/provenance/centernet.md in the LibreYOLO source repository.

Benchmarks

Independent accuracy and speed benchmarks: visionanalysis.org/model/centernet-resdcn18

License

The checkpoint publisher did not attach a standalone per-object license file. This mirror applies the releasing project's MIT license on an implied, not publisher-confirmed, basis. COCO annotations are CC BY 4.0; source images retain their individual Flickr terms. See LICENSE and NOTICE.