Research map / 3D vision / Face and shape
Robust Shape from Focus via Multiscale Directional Dilated Laplacian and Recurrent Network
Khurram Ashfaq, Muhammad Tariq Mahmood
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 |
|---|---|---|---|
| replace the DDL focus volume with a ConvNeXtV2-Atto encoder focus volume (ablation) | DDL + GRU pipeline · FT | MAE: 2.39 -> 2.03 | Improved |
| use DDL (0 params) instead of ResNet-18 encoder (11.2M params) for the focus volume (ablation) | GRU depth extractor · FT | Acc 1.25: 95.42 -> 95.73 | Improved |
| use DDL instead of VGG-16 encoder for the focus volume (ablation) | GRU depth extractor · FT | MAE: 8.99 -> 2.39 | Improved |
| use DDL instead of MobileNetV3 encoder for the focus volume (ablation) | GRU depth extractor · FT | MAE: 2.10 -> 2.39 | Hurt |
| use GRU recurrent depth extraction instead of AiFDNet regressor, same DDL focus volume (ablation) | DDL focus volume · FT | MAE: 5.45 -> 2.70 | Improved |
| use GRU recurrent depth extraction instead of DFV-FV regressor, same DDL focus volume (ablation) | DDL focus volume · FT | MSE: 370.50 -> 102.17 | Improved |
| add progressively larger dilation rates (G1 to G4) to DDL focus volume (ablation) | traditional SFF depth extraction · HCI/LFSD/FT under Gaussian, S&P, speckle noise | RMSE / CORR | Improved |
| add larger dilation rate (G2 over G1) to DDL in noise-free case (ablation) | traditional SFF depth extraction · HCI and FT, noise-free | RMSE / CORR | Hurt |
| add larger dilation rates to DDL in noise-free case on real data (ablation) | traditional SFF depth extraction · LFSD, noise-free | RMSE / CORR | Improved |
| use directional DDL G1 instead of 3x3 Laplacian (ablation) | traditional SFF focus measure · HCI/LFSD/FT noise conditions | RMSE / CORR | Improved |
| feed handcrafted DDL focus volume instead of raw focal stack or learnable focus volume (ablation) | AiFDNet · HCI | training convergence | Improved |
| increase GRU iterations from 1 to 4 (ablation) | DDL + GRU · FoD | inference time per stack (s): 0.0231 -> 0.0349 | Hurt |
Closest papers by meaning
- Deep Depth From Focus With Differential Focus Volume (CVPR 2022 · similarity 0.91)
- DualFocus: Depth from Focus with Spatio-Focal Dual Variational Constraints (NeurIPS 2025 · similarity 0.91)
- SAS: A General Framework Induced by Sequence Association for Shape From Focus (TPAMI 2025 · similarity 0.90)
- Noise Robust Depth From Focus Using a Ring Difference Filter (CVPR 2017 · similarity 0.90)
- Learning Depth from Focus in the Wild (ECCV 2022 · similarity 0.89)
- Multi-view Reconstruction via SfM-guided Monocular Depth Estimation (CVPR 2025 · similarity 0.86)
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.
