Research map / 3D vision / Face and shape
DiffBMP: Differentiable Rendering with Bitmap Primitives
Seongmin Hong, Junghun James Kim, Daehyeop Kim, Insoo Chung, Se Young Chun
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 soft rasterization (Gaussian blur) (ablation) | DiffBMP · Scenario 2 (512x512) | PSNR: 20.6 -> 21.5 | Improved |
| add structure-aware initialization (ablation) | DiffBMP · Scenario 3 (1024x1024) | PSNR: 25.9 -> 27.1 | Improved |
| combine soft rasterization and SA-init (ablation) | DiffBMP · Scenario 1 (512x512) | PSNR: 24.4 -> 25.7 | Improved |
| use custom CUDA kernel instead of PyTorch (ablation) | renderer · 1024x1024, RTX 3090 | backward runtime: 9811ms -> 5.8ms (CUDA-16bit) | Improved |
| use FP16 instead of FP32 CUDA (ablation) | renderer · 2048x2048, RTX 3090 | VRAM: 6.1 GB -> 3.8 GB | Improved |
| DiffBMP rendering vs DiffVG on complex SVG | rendering engine · Complex SVG primitives | PSNR: 14.35 dB -> 23.35 dB | Improved |
| initialize frame from previous frame's optimum (ablation) | Dynamic DiffBMP · video (17 videos) | tLP: 7.41 -> 5.39 | Improved |
| freeze primitives in unchanged regions (ablation) | Dynamic DiffBMP · video (17 videos) | tLP: 4.98 -> 3.49 | Improved |
| remove stuck primitives (ablation) | Dynamic DiffBMP · video (17 videos) | PSNR: 24.26 -> 24.66 | Improved |
| add all three video heuristics vs removing stuck only (ablation) | Dynamic DiffBMP · video (17 videos) | PSNR: 24.66 -> 24.38 | Mixed |
| uniform noise canvas (ablation) | DiffBMP · spatially constrained rendering | coverage | Improved |
Closest papers by meaning
- LinPrim: Linear Primitives for Differentiable Volumetric Rendering (NeurIPS 2025 · similarity 0.87)
- UniGS: Unified Geometry-Aware Gaussian Splatting for Multimodal Rendering (NeurIPS 2026 · similarity 0.86)
- FlashGS: Efficient 3D Gaussian Splatting for Large-scale and High-resolution Rendering (CVPR 2025 · similarity 0.86)
- Swift4D: Adaptive divide-and-conquer Gaussian Splatting for compact and efficient reconstruction of dynamic scene (ICLR 2025 · similarity 0.86)
- DashGaussian: Optimizing 3D Gaussian Splatting in 200 Seconds (CVPR 2025 · similarity 0.85)
- Gaussian Herding across Pens: An Optimal Transport Perspective on Global Gaussian Reduction for 3DGS (NeurIPS 2025 · 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.
