LibreSSD300

LibreSSD300

SSD300 with a VGG16 backbone, repackaged for LibreYOLO as an inference-only historic model. Input is fixed at 300 x 300.

from libreyolo import LibreYOLO

model = LibreYOLO("LibreSSD300.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: ssd300_vgg16_coco-b556d3b4.pth
SHA-256: b556d3b43ab6c3f63d81bfb8835fe8756ac22da664357da100dccf96b6a6b42d
Published COCO val2017 box mAP: 25.1.

VGG-16 initialization lineage

The torchvision SSD recipe initializes its backbone from VGG-16 feature weights released by the Visual Geometry Group, University of Oxford under CC BY 4.0. Attribution: Karen Simonyan and Andrew Zisserman, "Very Deep Convolutional Networks for Large-Scale Image Recognition," ICLR 2015.

Modifications

Torchvision modified the VGG graph for SSD and trained the detector on COCO. LibreYOLO adds v1.0 checkpoint metadata; learned tensors and state-dict keys are unchanged. The native graph has exact eager parity for preprocessing, both raw heads, default boxes, and final detections. ONNX Runtime prediction parity is also verified. See weights/convert_ssd_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 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. The Oxford attribution above records the VGG initialization lineage and does not claim that Oxford licensed the complete SSD checkpoint. See LICENSE and NOTICE.