LibreMaskRCNNr50

LibreMaskRCNNr50

Mask R-CNN with a ResNet-50-FPN v2 backbone, repackaged for LibreYOLO. The checkpoint supports instance segmentation by default and box-only detection with task="detect".

from libreyolo import LibreYOLO

model = LibreYOLO("LibreMaskRCNNr50.pt")
result = model.predict("image.jpg")
print(result.boxes.xyxy, result.masks.data)

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: maskrcnn_resnet50_fpn_v2_coco-73cbd019.pth

• Official SHA-256: 73cbd0190fcbe3ba339921fbce2c3a0b6bb9126c9a133c85e43a2a8e060a109e
• Converted SHA-256: 9214933a07cd354265e62c31298d4502f8433d124da6fea1b3c00cf78974cfbd
• Published COCO val2017 box mAP: 47.4
• Published COCO val2017 mask mAP: 41.8

Modifications

LibreYOLO checkpoint metadata was added. Learned tensors and state-dict keys are unchanged. The native graph loads the official state dict strictly and has exact eager parity at the RPN head, box head, final boxes, raw mask logits, and full-image masks. The batch-1 opset-18 ONNX graph is also covered by ONNX Runtime parity. See weights/convert_mask_rcnn_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.