Research map / 3D vision / Depth and stereo
TransNormal: Dense Visual Semantics for Diffusion-based Transparent Object Normal Estimation
Mingwei Li, Hehe Fan, Yi Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add edge-selective wavelet regularization (LL + edge HF) (ablation) | TransNormal · ClearPose | mean angular error (deg): 29.1 -> 25.5 | Improved |
| supervise only LL sub-band instead of LL plus edge HF (ablation) | TransNormal · ClearPose | mean angular error (deg): 25.5 -> 29.4 | Hurt |
| apply HF supervision on interior regions instead of edges (ablation) | TransNormal · ClearPose | mean angular error (deg): 25.5 -> 27.6 | Hurt |
| use DINOv3 instead of DINOv2 for semantic guidance (ablation) | TransNormal · ClearPose | mean angular error (deg): 28.5 -> 25.5 | Improved |
| use DINOv3 instead of SigLIP2 for semantic guidance (ablation) | TransNormal · ClearPose | mean angular error (deg): 27.2 -> 25.5 | Improved |
| use DINOv3 instead of SAM2 for semantic guidance (ablation) | TransNormal · ClearPose | mean angular error (deg): 28.5 -> 25.5 | Improved |
| add DINOv3 dense visual semantics via cross-attention (ablation) | TransNormal · ClearPose | mean angular error (deg): 27.7 -> 25.5 | Improved |
| fully fine-tune the U-Net instead of LoRA (ablation) | TransNormal · ClearPose | mean angular error (deg): 29.8 -> 25.5 | Improved |
| keep DINOv3 frozen instead of LoRA fine-tuning it (ablation) | TransNormal · ClearPose | mean angular error (deg): 27.5 -> 25.5 | Improved |
| replace text conditioning with DINOv3 visual semantics (TransNormal vs best prior) | diffusion-based normal estimator · ClearGrasp | mean angular error (deg): 21.7 -> 16.1 | Improved |
| replace text conditioning with DINOv3 visual semantics (TransNormal vs best prior) | diffusion-based normal estimator · ClearPose (real-world, zero-shot) | mean angular error (deg): 31.0 -> 25.5 | Improved |
Closest papers by meaning
- From Transparent to Opaque: Rethinking Neural Implicit Surfaces with $\alpha$-NeuS (NeurIPS 2024 · similarity 0.87)
- Normal-NeRF: Ambiguity-Robust Normal Estimation for Highly Reflective Scenes (AAAI 2025 · similarity 0.86)
- Neural LightRig: Unlocking Accurate Object Normal and Material Estimation with Multi-Light Diffusion (CVPR 2025 · similarity 0.86)
- Generative Perception of Shape and Material from Differential Motion (NeurIPS 2025 · similarity 0.85)
- Differentiable Neural Surface Refinement for Modeling Transparent Objects (CVPR 2024 · similarity 0.85)
- RichDreamer: A Generalizable Normal-Depth Diffusion Model for Detail Richness in Text-to-3D (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
