Research map / Architectures and efficiency / CNN design
Learning on the Edge: Investigating Boundary Filters in CNNs
Carlo Innamorati, Tobias Ritschel, Tim Weyrich, Niloy J. Mitra
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 explicit boundary filters instead of zero padding | U-net-like CNN · Gauss filtering | DSSIM: .0022 -> .0016 | Improved |
| use explicit boundary filters instead of reflect padding | U-net-like CNN · Gauss filtering | DSSIM: .0018 -> .0016 | Improved |
| use explicit boundary filters instead of zero padding | encoder-decoder · De-noise/Bayer | PSNR: 31.50 -> 31.94 | Improved |
| use explicit boundary filters instead of zero padding | encoder-decoder · Colorization | DSSIM: .1604 -> .1577 | Improved |
| use explicit boundary filters instead of zero padding | encoder-decoder · Disparity | pixel error: 1.511 -> 1.403 | Improved |
| use explicit boundary filters instead of zero padding | encoder-decoder · Scene flow | pixel error: 1.183 -> 1.096 | Improved |
| use explicit boundary filters instead of reflect padding | encoder-decoder · Scene flow | MSE ratio: 56% -> 100% | Improved |
| use explicit boundary filters instead of zero/reflect while varying network depth and width (ablation) | simple Gaussian blur CNN · Gaussian blur learning | MSE | Improved |
| replace nine-convolution implementation with domain decomposition (ablation) | explicit boundary filters · Boundary overhead | runtime: 22.5 -> 2.5 ms | Improved |
| replace per-pixel conditional kernel select with domain decomposition (ablation) | explicit boundary filters · Boundary overhead | runtime: 11.2 -> 2.5 ms | Improved |
| add explicit boundary filters in Keras training (ablation) | encoder-decoder · Training time per epoch | performance loss: 0.2% | No clear effect |
Closest papers by meaning
- Learning to Predict Crisp Boundaries (ECCV 2018 · similarity 0.91)
- Deep Edge Filter: Return of the Human-Crafted Layer in Deep Learning (NeurIPS 2025 · similarity 0.89)
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.89)
- Convolutional Oriented Boundaries: From Image Segmentation to High-Level Tasks (TPAMI 2017 · similarity 0.89)
- Semantic Segmentation With Boundary Neural Fields (CVPR 2016 · similarity 0.88)
- RePr: Improved Training of Convolutional Filters (CVPR 2019 · 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.
