Research map / Learning theory and algorithms / Combinatorial and search optimization
Efficient Continuous Relaxations for Dense CRF
Alban Desmaison, Rudy Bunel, Pushmeet Kohli, Philip H. S. Torr, M. Pawan Kumar
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 |
|---|---|---|---|
| use continuous relaxation (LP) instead of mean-field inference for dense CRF energy minimisation | dense CRF · PASCAL VOC 2010 (params tuned for DC_neg) | average final energy: -600 -> -6697 | Improved |
| use DC_neg relaxation instead of mean-field (MF) inference | dense CRF · PASCAL VOC 2010 (params tuned for DC_neg) | IoU: 27.53 -> 28.60 | Improved |
| use LP relaxation instead of DC_neg | dense CRF · PASCAL VOC 2010 (params tuned for DC_neg) | average final energy: -6613 -> -6697 | Improved |
| run mean-field to convergence (MF) instead of 5 iterations (MF5) | mean-field · PASCAL VOC 2010 (params tuned for DC_neg) | average final energy: -600 -> -600 | No clear effect |
| use continuous relaxations (QP_cvx) with parameters tuned for MF instead of mean-field | dense CRF · PASCAL VOC 2010 (params tuned for MF) | IoU: 28.86 -> 14.94 | Hurt |
| use QP_cvx instead of MF with parameters tuned for MF | dense CRF · PASCAL VOC 2010 (params tuned for MF) | average final energy: -1.19e6 -> -1.66e6 | Improved |
| tune pixel compatibility parameters for DC_neg instead of MF (using LP) (ablation) | LP relaxation · PASCAL VOC 2010 | IoU: 14.93 -> 28.68 | Improved |
| use DC_neg relaxation instead of mean-field on stereo matching | dense CRF · Middlebury stereo | final energy | Improved |
Closest papers by meaning
- Efficient Linear Programming for Dense CRFs (CVPR 2017 · similarity 0.91)
- Regularized Frank-Wolfe for Dense CRFs: Generalizing Mean Field and Beyond (NeurIPS 2021 · similarity 0.91)
- Efficient Graph Cut Optimization for Full CRFs with Quantized Edges (TPAMI 2019 · similarity 0.89)
- Sparse Non-Local CRF With Applications (TPAMI 2024 · similarity 0.88)
- Dense and Low-Rank Gaussian CRFs Using Deep Embeddings (ICCV 2017 · similarity 0.88)
- Convex Relaxations for Manifold-Valued Markov Random Fields with Approximation Guarantees (ECCV 2024 · similarity 0.88)
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.
