# Third-party notices — LRASPP MobileNetV3-Large

This directory contains a self-exported ONNX version of torchvision's pretrained
**LRASPP MobileNetV3-Large** semantic segmentation model, used in place of the
heavier FCN-ResNet50 model to reduce in-browser inference latency.

- Source repository: https://github.com/pytorch/vision
- Pretrained weights: https://download.pytorch.org/models/lraspp_mobilenet_v3_large-d234d4ea.pth
  (`torchvision.models.segmentation.LRASPP_MobileNet_V3_Large_Weights.DEFAULT`)
- License: BSD-3-Clause (verified against `pytorch/vision`'s `LICENSE` file — see `./LICENSE`
  in this directory) — **commercial use permitted**.
- Export: converted from the pretrained PyTorch weights to ONNX (opset 17, single-file) via
  `torch.onnx.export`, with no changes to model weights or architecture.
- Params: 3,221,538 (~11x fewer than the FCN-ResNet50 model it replaces).
- Trained on the same 21-class Pascal VOC label set (`__background__`, `aeroplane`, ...,
  `tvmonitor`) as the model it replaces, so no changes were needed to the site's class
  list, color palette, or output decoding.

No modifications were made to the model weights or architecture beyond the ONNX export
itself.