Update README.md

README.md

@@ -3,46 +3,46 @@ license: mit
 language:
 - en
 tags:
-- 3D
-- 3D-Reconstruction
-- Sketch-to-3D
-- Transformer
-- Pytorch
+- 3d
+- 3d-reconstruction
+- sketch-based-modeling
+- surface-reconstruction
+- pytorch
+- siggraph-2026
 ---
 
-# S2V-Net (NeuralSketch2Surf)
+# S2V-Net for NeuralSketch2Surf
 
-S2V-Net is the core model from the paper **"NeuralSketch2Surf: Fast Neural Surfacing of Unoriented 3D Sketches"**. It instantly converts sparse, unoriented 3D sketches (like those drawn in VR) into smooth, closed 3D meshes.
+This repository hosts the pretrained S2V-Net weights used by **NeuralSketch2Surf: Fast Neural Surfacing of Unoriented 3D Sketches**.
 
-![Teaser Image](./Architecture.png)
-![result Image](./UpdatedResults.png)
-![datasets Image](./UpdatedDataset.png)
+S2V-Net reconstructs closed surfaces from sparse, unoriented 3D sketch curves. The model predicts a `112^3` volumetric occupancy field from voxelized sketch strokes; the final mesh is extracted with Marching Cubes and can be refined with the smoothing tool from the project repository.
 
-## Core Highlights
+![Architecture](./Architecture.png)
+![Results](./UpdatedResults.png)
+![Cross-dataset results](./UpdatedDataset.png)
 
-*   **Lightning Fast (Interactive Rates):** The entire pipeline takes **< 0.7 seconds** on a GPU, making it perfectly suited for real-time VR applications.
-*   **No Normals Required:** Unlike many existing methods, it processes completely unoriented, raw 3D strokes. Users don't need to worry about stroke directions.
-*   **Robust & Accurate:** Accurately fills large spatial gaps between sparse strokes while preserving high-frequency geometric details. 
+## Model
 
-## How it Works
+- **Input:** voxelized 3D sketch strokes on a `112^3` grid.
+- **Output:** binary surface occupancy probabilities.
+- **Backbone:** SwinUNETR-style 3D transformer for global shape inference.
+- **Refinement:** lightweight 3D residual module for local boundary correction.
+- **Use case:** interactive sketch-based surface reconstruction from raw 3D strokes.
 
-The pipeline treats 3D surfacing as a binary voxel-occupancy prediction task operating on a $112^3$ grid:
-1. **Backbone (Global Shape):** A custom **SwinUNETR v2** transformer infers the global topology and bridges large gaps between sparse input strokes.
-2. **Refinement (Local Details):** A lightweight 3D CNN acts as a geometric denoiser to sharpen boundaries and recover fine details.
-3. **Meshing:** The predicted occupancy grid is extracted via Marching Cubes and smoothed using a locally controllable Laplacian filter.
+## Notes
 
-## Limitations
-*   **Closed Surfaces Only:** The model assumes the input sketch represents a solid, closed object. It is not designed for open surfaces.
+- The model is trained for closed-surface reconstruction.
+- Very thin structures may be limited by the `112^3` voxel resolution.
 
 ## Citation
 
-If you use S2V-Net in your research or project, please cite:
+If you use these weights, please cite:
 
 ```bibtex
-@inproceedings{neuralsketch2surf2026,
-  author = {Anonymous Author(s)},
-  title = {NeuralSketch2Surf: Fast Neural Surfacing of Unoriented 3D Sketches},
-  booktitle = {Proceedings of ACM Trans. Graph.},
-  year = {2026},
-  publisher = {ACM}
-}
+@article{neuralsketch2surf2026,
+  title   = {NeuralSketch2Surf: Fast Neural Surfacing of Unoriented 3D Sketches},
+  author  = {Ye, Hongsheng and Sureshkumar, Anandhu and Wang, Zhonghan and Cani, Marie-Paule and Hahmann, Stefanie and Bonneau, Georges-Pierre and Parakkat, Amal Dev},
+  journal = {ACM Transactions on Graphics (Proceedings of SIGGRAPH)},
+  year    = {2026}
+}
+```