Research map / 3D vision / Depth and stereo
2K Retrofit: Entropy-Guided Efficient Sparse Refinement for High-Resolution 3D Geometry Prediction
Tianbao Zhang, Zhenyu Liang, Zhenbo Song, Nana Wang, Xiaomei Zhang, Xudong Cai, Zheng Zhu, Kejian Wu, Gang Wang, Zhaoxin Fan
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 |
|---|---|---|---|
| select pixels by entropy instead of randomly (ablation) | 2K Retrofit · ETH3D point map | Accuracy error: 1.126 -> 0.935 | Improved |
| use entropy selector instead of edge-based heuristics (ablation) | 2K Retrofit · ETH3D point map | Accuracy error: 0.942 -> 0.935 | Improved |
| use learnable selector instead of entropy selector (ablation) | 2K Retrofit · ETH3D point map | FPS: 5.5 -> 3.8 | Hurt |
| use gated fusion instead of direct replacement (ablation) | 2K Retrofit · ETH3D point map | Accuracy error: 1.113 -> 0.935 | Improved |
| use gated fusion instead of entropy-based fusion (ablation) | 2K Retrofit · ETH3D point map | Accuracy error: 0.940 -> 0.935 | Improved |
| lower entropy threshold alpha from 0.3 to 0.1 (ablation) | 2K Retrofit · ETH3D point map | Accuracy error / FPS: Acc 0.935 -> 0.917; FPS 5.5 -> 3.4 | Mixed |
| raise entropy threshold alpha from 0.3 to 0.8 (ablation) | 2K Retrofit · ETH3D point map | Accuracy error / FPS: Acc 0.935 -> 1.135; FPS 5.5 -> 8.0 | Mixed |
| use Point Transformer instead of MinkowskiUNet as sparse extractor (ablation) | 2K Retrofit · ETH3D point map | Accuracy error / FPS: Acc 0.935 -> 0.919; FPS 5.5 -> 1.7 | Mixed |
| use 2K Retrofit instead of retraining VGGT at 2K | VGGT · ETH3D point map | Accuracy error: 0.911 -> 0.935 | Hurt |
| use 2K Retrofit instead of retraining VGGT at 2K | VGGT · ETH3D point map | GFLOPs: 495 -> 172 | Improved |
| add 2K Retrofit sparse refinement to VGGT | VGGT · ETH3D point map | Overall error: 0.903 -> 0.839 | Improved |
| use 2K Retrofit instead of PRO for monocular depth | PRO · ETH3D monocular 2K depth | AbsRel: 0.0217 -> 0.0192 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VGGT — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Leveraging VLM-Based Pipelines to Annotate 3D Objects (ICML 2024 · similarity 0.84)
- Twin-T & TwintVQA: A Reliable Structure-Detail Separating VLM and a Comprehensive Benchmark for Chart and Table Tasks (CVPR 2026 · similarity 0.84)
- S$^2$-MLLM: Boosting Spatial Reasoning Capability of MLLMs for 3D Visual Grounding with Structural Guidance (CVPR 2026 · similarity 0.84)
- G$^2$VLM: Geometry Grounded Vision Language Model with Unified 3D Reconstruction and Spatial Reasoning (CVPR 2026 · similarity 0.84)
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026 · similarity 0.84)
- LayoutVLM: Differentiable Optimization of 3D Layout via Vision-Language Models (CVPR 2025 · similarity 0.83)
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.
