Research map / 3D vision / Face and shape
InstantSticker: Realistic Decal Blending via Disentangled Object Reconstruction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes InstantSticker, a disentangled IBL-based mesh reconstruction pipeline with ARAP-based pre-parameterized UV area, neural albedo texture, and shadow factor that enables instant decal blending and real-time rendering.
Problem
Realistic decal blending on reconstructed 3D objects suffers from warping, blurriness, slow editing latency, and baked-in shadows in prior NeRF-based methods.
Approach
Combines surface mesh rasterization with IBL and Disney BRDF, pre-parameterizes a selected mesh area via ARAP with a neural (MLP-generated) albedo texture, and adds a learnable shadow factor so shadows are decomposed from albedo, allowing instant albedo texture replacement for decal editing.
Key results
- Ours achieves 28.81 PSNR on Shiny Blender and 26.12 on DTU, comparable to DE-NeRF (29.18 NeRF Syn., 28.79 Shiny, 25.93 DTU).
- Rendering reaches 81.5 FPS versus 0.013-0.017 FPS for NeuMesh and DE-NeRF, with zero editing time and warping variance 0.001 vs 0.472-0.565 for baselines.
- User study with 18 participants gives a realism score of 0.86 versus 0.55 for DE-NeRF.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use InstantSticker instead of DE-NeRF | Decal blending pipeline · Decal blending | RVW: 0.489 -> 0.001 | Helped |
| Use InstantSticker instead of DE-NeRF | Decal blending pipeline · Decal blending | FPS: 0.013 -> 81.5 | Helped |
| Use InstantSticker instead of DE-NeRF | Decal blending pipeline · Decal blending | editing time: 1 hour -> 0 | Helped |
| Use InstantSticker instead of DE-NeRF | Decal blending pipeline · User study | preference score: 0.86 vs. 0.55 | Helped |
| Add pre-parameterization before appearance disentangling instead of post-parameterization (ablation) | InstantSticker · Decal appearance | high-frequency detail | Helped |
| Replace pixel-optimized texture maps with an MLP neural texture map (ablation) | InstantSticker · Novel views | texture rendering quality | Helped |
| Add the local reflection network (ablation) | InstantSticker · Specular object reconstruction | local reflection reconstruction | Helped |
| Add the shadow factor (ablation) | InstantSticker · Decal blending | shadow realism | Helped |
Limitations
- Depends on high-quality mesh input.
- Pre-parameterization applies only to a user-selected UV area rather than whole object.
Experimental setup
Datasets: NeRF-Synthetic Shiny Blender DTU
Benchmarks: NeRF-Synthetic Shiny Blender DTU
Metrics: PSNR FPS editing time training time Ratio Variance Warping (RVW) user preference score
Baselines: NeuMesh Seal-3D DE-NeRF PhySG NeRFactor NvDiffRec
Builds on: Image-Based Lighting Disney BRDF DE-NeRF NeuMesh Voxurf Ref-NeuS Ref-NeRF
Closest papers by meaning
- Factored-NeuS: Reconstructing Surfaces, Illumination, and Materials of Possibly Glossy Objects (CVPR 2025 · similarity 0.85)
- RNG: Relightable Neural Gaussians (CVPR 2025 · similarity 0.84)
- Eclipse: Disambiguating Illumination and Materials using Unintended Shadows (CVPR 2024 · similarity 0.84)
- Inverse Rendering of Glossy Objects via the Neural Plenoptic Function and Radiance Fields (CVPR 2024 · similarity 0.84)
- Textured Gaussians for Enhanced 3D Scene Appearance Modeling (CVPR 2025 · similarity 0.84)
- ReGS: Reference-based Controllable Scene Stylization with Gaussian Splatting (NeurIPS 2024 · similarity 0.84)
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.