Research map / Detection and segmentation / CNN features
Higher Order Conditional Random Fields in Deep Neural Networks
Anurag Arnab, Sadeep Jayasumana, Shuai Zheng, Philip H. S. Torr
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 |
|---|---|---|---|
| add detection higher order potentials (ablation) | Pairwise CRF-RNN · VOC 2012 reduced val | mean IoU (%): 72.9 -> 74.9 | Improved |
| add superpixel higher order potentials (ablation) | Pairwise CRF-RNN · VOC 2012 reduced val | mean IoU (%): 72.9 -> 74.0 | Improved |
| add both detection and superpixel potentials (ablation) | Pairwise CRF-RNN · VOC 2012 reduced val | mean IoU (%): 72.9 -> 75.8 | Improved |
| add superpixel potentials on top of detection potentials (ablation) | Detections-only CRF · VOC 2012 reduced val | mean IoU (%): 74.9 -> 75.8 | Improved |
| train CRF end-to-end instead of piecewise (ablation) | Higher Order CRF (FCN-8s) · VOC 2012 reduced val | mean IoU (%): 73.6 -> 75.8 | Improved |
| train CRF end-to-end instead of piecewise (ablation) | Higher Order CRF (DCN) · VOC 2012 reduced val | mean IoU (%): 73.5 -> 75.0 | Improved |
| train pairwise CRF end-to-end instead of piecewise (ablation) | Pairwise CRF (FCN-8s) · VOC 2012 reduced val | mean IoU (%): 69.5 -> 72.9 | Improved |
| add higher order potentials to pairwise CRF with DCN unaries | Pairwise CRF (DCN) · VOC 2012 reduced val | mean IoU (%): 72.5 -> 75.0 | Improved |
| use latent-variable detection potentials instead of adding detection mask to unaries (ablation) | Detection unary baseline · VOC 2012 reduced val | mean IoU (%): 73.6 -> 74.9 | Improved |
| add higher order potentials to CRF-RNN | CRF-RNN · VOC 2012 test | mean IoU (%): 74.7 -> 77.9 | Improved |
| add higher order potentials to CRF-RNN | CRF-RNN · PASCAL Context val (59 classes) | mean IoU (%): 39.3 -> 41.3 | Improved |
| add superpixel potentials (ablation) | Pairwise CRF-RNN · VOC 2012 reduced val (boundary IoU) | boundary IoU | Hurt |
Closest papers by meaning
- Gaussian Conditional Random Field Network for Semantic Segmentation (CVPR 2016 · similarity 0.92)
- Fast, Exact and Multi-scale Inference for Semantic Image Segmentation with Deep Gaussian CRFs (ECCV 2016 · similarity 0.91)
- E-CRF: Embedded Conditional Random Field for Boundary-caused Class Weights Confusion in Semantic Segmentation (ICLR 2023 · similarity 0.89)
- Regularized Frank-Wolfe for Dense CRFs: Generalizing Mean Field and Beyond (NeurIPS 2021 · similarity 0.89)
- Efficient Linear Programming for Dense CRFs (CVPR 2017 · similarity 0.88)
- Dense and Low-Rank Gaussian CRFs Using Deep Embeddings (ICCV 2017 · 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.
