Research map / Detection and segmentation / CNN features
Dynamic Filtering with Large Sampling Field for ConvNets
Jialin Wu, Dai Li, Yu Yang, Chandrajit Bajaj, Xiangyang Ji
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 an LS-DFN layer | R-FCN · VOC 2007 detection | mAP: 77.6 -> 79.2 | Improved |
| add an LS-DFN layer | CoupleNet · VOC 2007 detection | mAP: 80.4 -> 81.7 | Improved |
| sample 3x3 neighbour regions instead of only the identical position (s=1 to s=3) (ablation) | LS-DFN, C'=24, with attention · VOC 2007 detection | mAP: 72.5 -> 78.6 | Improved |
| increase samples from s=3 to s=5 (ablation) | LS-DFN, C'=24, with attention · VOC 2007 detection | mAP: 78.6 -> 78.6 | No clear effect |
| add attention mechanism (stride 1) (ablation) | LS-DFN, C'=24 · VOC 2007 detection | mAP: 77.4 -> 78.1 | Improved |
| add attention mechanism (C'=32, stride 1) (ablation) | LS-DFN · VOC 2007 detection | mAP: 77.6 -> 78.6 | Improved |
| add residual learning to kernel generation (ablation) | LS-DFN, C'=24, with attention · VOC 2007 detection | mAP: 68.1 -> 78.68 | Improved |
| add residual learning without attention (ablation) | LS-DFN, C'=24 · VOC 2007 detection | convergence: fails to converge -> 77.4 | Improved |
| increase stride from 1 to 2 (C'=24, with attention) (ablation) | LS-DFN · VOC 2007 detection | mAP: 78.1 -> 78.6 | Improved |
| increase stride from 1 to 2 (C'=32, with attention) (ablation) | LS-DFN · VOC 2007 detection | mAP: 78.6 -> 78.0 | Hurt |
| add LS-DFN layer to DeepLabV2 + CRF | DeepLabV2 · VOC 2012 segmentation | mean IoU: 79.7 -> 81.1 | Improved |
| replace dilated convolutions by LS-DFN with strides | DeepLabV2 · VOC 2012 segmentation | mean IoU: 79.7 -> 80.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CoupleNet — CoupleNet: Coupling Global Structure With Local Parts for Object Detection (ICCV 2017) · better on 2 of 2 shared results
- R-FCN — R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016) · 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
- Dynamic Filter Networks (NeurIPS 2016 · similarity 0.88)
- UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition (CVPR 2024 · similarity 0.85)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.85)
- Frequency Dynamic Convolution for Dense Image Prediction (CVPR 2025 · similarity 0.84)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.84)
- CAM Back Again: Large Kernel CNNs from a Weakly Supervised Object Localization Perspective (CVPR 2024 · similarity 0.84)
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.
