Viscan OPG - Faster R-CNN ResNet-50 FPN V2 Weights
Fine-tuned weights for automated pathology detection in dental panoramic radiographs (OPG) on the DENTEX (MICCAI 2023) dataset.
Model Summary
- Architecture: Faster R-CNN with ResNet-50 FPN V2 backbone
- Pretrained Weights: COCO_V1
- Classes (5): ackground, impacted_tooth, caries, periapical_lesion, deep_caries
- Total Parameters: 43,271,528 (43,046,184 trainable)
- File Size: 173,466,789 bytes
- SHA-256: e0b49b12f48fc71ec1f583db08f107650659571e7c2b9cbd454dbfed13b46a1f
- Operating Confidence Threshold: 0.70 (tuned via validation macro-F1 sweep)
License Notice
- Model Code: BSD-3-Clause (torchvision)
- Model Weights: CC BY-NC-SA 4.0 (derived from fine-tuning on DENTEX MICCAI benchmark). Non-commercial academic and research use.