Research map / Architectures and efficiency / CNN design
Fast End-to-End Trainable Guided Filter
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 end-to-end training of the guided filtering layer rather than train the low-resolution network separately and use a fixed filter (ablation) | DGF on non-local dehazing · MIT-Adobe FiveK | PSNR: 1 dB | Helped |
| Add end-to-end training of the guided filtering layer rather than train the low-resolution network separately and use a fixed filter (ablation) | DGF on detail manipulation · MIT-Adobe FiveK | PSNR: 0.83 dB | Helped |
| Replace non-parametric filtering operations with learnable convolution layers (ablation) | DGF_b · Five image processing tasks | PSNR | Helped |
| Add a learnable task-specific guidance transformation F(I) (ablation) | DGF_b^c · Style Transfer | PSNR: 2.56 dB | Helped |
| Add a learnable task-specific guidance transformation F(I) (ablation) | DGF_b^c · Detail Manipulation | PSNR: 1.62 dB | Helped |
| Increase the guided-filter radius r above the default (ablation) | DGF guided filtering layer · Image processing tasks | PSNR | Hurt |
| Increase the low-resolution input I_l resolution (ablation) | DGF · L0 Smoothing, Detail Manipulation, and Non-local Dehazing | PSNR | Helped |
| Increase the low-resolution input I_l resolution (ablation) | DGF · Style Transfer and Image Retouching | PSNR | Mixed |
| Increase the dilation rate of F(I) (ablation) | DGF^c · Image processing tasks | PSNR | Helped |
| Apply the guided filtering layer to the saliency baseline (ablation) | DSS · MSRA-B saliency detection | F_beta: 90.61% to 91.29% | Helped |
| Replace DGF_s with end-to-end trainable DGF (ablation) | Saliency detection · MSRA-B | F_beta: 91.29% to 91.75% | Helped |
| Replace the baseline segmentation method with DGF | DeepLab-V2 · PASCAL VOC 2012 validation | Mean IOU: 1.79% | Helped |
Closest papers by meaning
- Spatially Variant Linear Representation Models for Joint Filtering (CVPR 2019 · similarity 0.87)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.86)
- Learning Versatile Filters for Efficient Convolutional Neural Networks (NeurIPS 2018 · similarity 0.86)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.85)
- Deep Edge Filter: Return of the Human-Crafted Layer in Deep Learning (NeurIPS 2025 · similarity 0.85)
- The Master Key Filters Hypothesis: Deep Filters Are General (AAAI 2025 · 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.