Research map / Detection and segmentation / CNN features
Learnable Histogram: Statistical Context Features for Deep Neural Networks
Zhe Wang, Hongsheng Li, Wanli Ouyang, Xiaogang Wang
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 the learnable histogram layer (HistNet-SS) to FCN | VGG-FCN · SIFTFlow | per-pixel accuracy: 0.860 -> 0.879 | Improved |
| add the learnable histogram layer (HistNet-SS) to FCN | VGG-FCN · Stanford background | per-pixel accuracy: 0.851 -> 0.871 | Improved |
| add the learnable histogram layer (HistNet-OD) to faster-RCNN | faster RCNN · VOC 2007 test | mAP: 69.5 -> 71.4 | Improved |
| add the learnable histogram layer to DeepLab | DeepLab · VOC 2012 segmentation | mean IOU: 64.2 -> 67.5 | Improved |
| learn bin centers and widths instead of fixing them (ablation) | HistNet-SS · SIFTFlow | per-pixel accuracy: 0.872 -> 0.879 | Improved |
| lock histogram kernels to keep physical meaning instead of unlocking all (ablation) | HistNet-SS · SIFTFlow | per-pixel accuracy: 0.870 -> 0.879 | Improved |
| use histogram (statistical) context instead of global-pooled fc7 features (ablation) | HistNet-SS · SIFTFlow | per-pixel accuracy: 0.870 -> 0.879 | Improved |
| use histogram context instead of global-pooled score features (ablation) | HistNet-SS · SIFTFlow | per-pixel accuracy: 0.873 -> 0.879 | Improved |
| use histogram context instead of fc7 global features (ablation) | HistNet-OD · VOC 2007 test | mAP: 0.704 -> 0.714 | Improved |
| learn histogram parameters instead of fixing them (ablation) | HistNet-OD · VOC 2007 test | mAP: 0.707 -> 0.714 | Improved |
| add a second recurrent histogram layer (R-HistNet-SS) (ablation) | HistNet-SS · SIFTFlow | per-pixel accuracy: 0.879 -> 0.880 | No clear effect |
| add fully connected CRF post-processing | HistNet-SS · SIFTFlow | per-class accuracy: 0.5 -> 0.512 | 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.
- faster RCNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · better on 1 of 1 shared result
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
- Contextual Convolutional Networks (ICLR 2023 · similarity 0.83)
- Investigating the Histogram Loss in Regression (JMLR 2026 · similarity 0.83)
- Learning Deep Embeddings with Histogram Loss (NeurIPS 2016 · similarity 0.82)
- Fully Convolutional Networks for Semantic Segmentation (TPAMI 2016 · similarity 0.82)
- Region-Based Convolutional Networks for Accurate Object Detection and Segmentation (TPAMI 2016 · similarity 0.82)
- Towards flexible perception with visual memory (ICML 2025 · similarity 0.82)
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.
