Research map / Detection and segmentation / Open-set and few-shot detection
Low Compute and Fully Parallel Computer Vision With HashMatch
Sean Ryan Fanello, Julien Valentin, Adarsh Kowdle, Christoph Rhemann, Vladimir Tankovich, Carlo Ciliberto, Philip Davidson, Shahram Izadi
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 proposed parallel inference on top of the hash-code data term (ablation) | HashMatch · Synthetic stereo disparity | accuracy (% pixels below 1 px error): 77% -> 96% | Improved |
| use the proposed parallel inference instead of PatchMatch inference with the same unary term (ablation) | HashMatch depth estimation · Active stereo depth | speed: 2X faster | Improved |
| add inference propagation over the unaries alone (ablation) | Background subtraction CRF · Background subtraction | average energy: 5.8 ∗105 -> 2.9 ∗105 | Improved |
| use the proposed parallel inference instead of TRW | Background subtraction CRF · Background subtraction | average energy: 2.85 ∗105 -> 2.9 ∗105 | No clear effect |
| use sparse HashMatch codes (4 non-zeros) instead of dense Binary Autoencoders | hash codes for stereo matching · Synthetic stereo disparity | accuracy: 77% -> 77% | No clear effect |
| use HashMatch codes instead of other sparse hashing schemes (CBE, SBE) | hash codes for stereo matching · Synthetic stereo disparity | accuracy: 70% -> 77% | Improved |
| approximate SIFT with HashMatch descriptors | wide-baseline matching · EPFL wide-baseline stereo | match inlier accuracy: 81% -> 90% | Improved |
| approximate SIFT with HashMatch descriptors | wide-baseline matching · EPFL wide-baseline stereo | average end-point error: 0.8 cm -> 1.2 cm | Hurt |
| approximate SIFT with HashMatch descriptors | wide-baseline matching · EPFL wide-baseline stereo | good matches per image: 350 -> 150 | Hurt |
| use HashMatch codes instead of dense and sparse baselines at small code lengths | GIST1M nearest neighbor retrieval · GIST1M | Recall@R | Mixed |
Closest papers by meaning
- Detect, Replace, Refine: Deep Structured Prediction for Pixel Wise Labeling (CVPR 2017 · similarity 0.86)
- PatchMatch Filter: Edge-Aware Filtering Meets Randomized Search for Visual Correspondence (TPAMI 2016 · similarity 0.85)
- On the Convergence of PatchMatch and Its Variants (CVPR 2018 · similarity 0.84)
- On the Synergies between Machine Learning and Binocular Stereo for Depth Estimation from Images: a Survey (TPAMI 2021 · similarity 0.84)
- Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches (JMLR 2016 · similarity 0.84)
- RUBIK: A Structured Benchmark for Image Matching across Geometric Challenges (CVPR 2025 · similarity 0.83)
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.
