TEED, exported to ONNX

TEED (Tiny and Efficient Edge Detector, 58 K parameters) by Xavier Soria Poma et al., converted to ONNX for TerraGen. MIT, as upstream (LICENSE is upstream's, unchanged).

  • Weights: checkpoints/BIPED/5/5_model.pth (upstream's default checkpoint) at xavysp/TEED commit 40fa4b1391dc6424f88989d0ca75d5b592c8681d. The authors do not state what the released weights were trained on beyond the BIPED checkpoint folder's name.
  • Input: image float32 [1, 3, H, W], BGR order, 0-255, with the mean (104.007, 116.669, 122.679) subtracted; H and W multiples of 8.
  • Output: edges float32 [1, 1, H, W], edge probability 0-1 (sigmoid of TEED's fused output).
  • Checked against PyTorch on random input: largest difference 2.6e-6.

Citation: Soria, Li, Rouhani, Sappa, "Tiny and Efficient Model for the Edge Detection Generalization", ICCV 2023 Workshops (arXiv:2308.06468).

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Paper for Shaboomi/teed-onnx