LibreRetinaNetr50

RetinaNet (ResNet-50 FPN v1 with FrozenBatchNorm), repackaged for LibreYOLO. This is an inference-only model with 34,014,999 parameters.

from libreyolo import LibreYOLO

model = LibreYOLO("LibreRetinaNetr50.pt")
results = model.predict("image.jpg")

Source

Derived from pytorch/vision at commit 336d36e8db990a905498c73933e35231876e28bc. Copyright (c) Soumith Chintala 2016 and torchvision contributors. The source implementation is BSD-3-Clause.

Official checkpoint: retinanet_resnet50_fpn_coco-eeacb38b.pth

• Official file bytes: 136595076
• Official SHA-256: eeacb38b7cec8cf93c57867e05eaab621047f19b0d2ec5accaa405f690da15b7
• Converted file bytes: 136594812
• Converted SHA-256: a2b9d711f531bbee88eff659d11ca263792491060fa6fe0ab957541b63f12051
• Published COCO val2017 box mAP: 36.4

Model contract

• Input: RGB image, normalized with ImageNet mean/std.
• Resize: short side 800, long side capped at 1333, then bottom/right padding to a multiple of 32.
• Output: contiguous COCO-80 boxes, scores, and class ids after per-level candidate selection and class-aware NMS.
• Training: not implemented in LibreYOLO; train() raises.
• Export: dynamic-spatial, batch-one ONNX is validated.

Modifications

Checkpoint metadata was added for LibreYOLO's v1.0 schema. Learned tensors and state-dict keys are unchanged. The native LibreYOLO graph strictly loads the official state dict and has exact eager parity at every FPN feature, raw head, and final detection. See weights/convert_retinanet_weights.py in the LibreYOLO source repository.

License

The checkpoint publisher did not attach a separate per-object license file. This mirror applies the releasing project's BSD-3-Clause license on an implied, not publisher-confirmed, basis. Torchvision warns that pretrained models may have their own licenses or terms derived from training data and that users must determine whether they have permission for their use case. COCO annotations are CC BY 4.0; source images retain their individual Flickr terms. See LICENSE and NOTICE.