Research map / 3D vision / Depth and stereo
Warp Consistency for Unsupervised Learning of Dense Correspondences
Prune Truong, Martin Danelljan, Fisher Yu, Luc Van Gool
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 warp-supervision term to W-bipath loss (ablation) | W-bipath GLU-Net · MegaDepth | PCK-1: 29.55 -> 39.66 | Improved |
| stop gradients through the flow used for warping (ablation) | W-bipath GLU-Net · MegaDepth | PCK-1: 20.06 -> 29.55 | Improved |
| add adaptive loss balancing instead of lambda=1 (ablation) | WarpC GLU-Net · MegaDepth | PCK-1: 39.66 -> 43.47 | Mixed |
| add visibility mask in second stage (ablation) | WarpC GLU-Net · MegaDepth | PCK-1: 43.47 -> 48.86 | Improved |
| sample harder warps with elastic transforms (ablation) | WarpC GLU-Net · MegaDepth | PCK-1: 48.86 -> 50.61 | Improved |
| use W-bipath instead of warp-supervision alone (ablation) | GLU-Net · MegaDepth | PCK-5: 57.21 -> 67.70 | Improved |
| combine W-bipath with warp-supervision (ablation) | GLU-Net · MegaDepth | PCK-5: 67.70 -> 69.90 | Improved |
| use I'J-bipath constraint as loss (ablation) | GLU-Net · MegaDepth | PCK-5: 0.05 | Hurt |
| use WarpC instead of SSIM loss (ablation) | GLU-Net · MegaDepth | PCK-5: 69.58 -> 78.61 | Improved |
| use WarpC unsupervised instead of sparse ground-truth supervision (ablation) | GLU-Net · MegaDepth | PCK-5: 72.42 -> 78.61 | Improved |
| fine-tune WarpC with sparse annotations (ablation) | WarpC GLU-Net · MegaDepth vs RobotCar | PCK-5: 78.61 -> 81.33 (MegaDepth); 35.92 -> 34.67 (RobotCar) | Mixed |
| add SSIM photometric loss to WarpC (ablation) | WarpC GLU-Net · MegaDepth | PCK-1: 50.61 -> 54.92 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RANSAC-Flow — RANSAC-Flow: Generic Two-stage Image Alignment (ECCV 2020) · 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
- Probabilistic Warp Consistency for Weakly-Supervised Semantic Correspondences (CVPR 2022 · similarity 0.91)
- Deep Matching Prior: Test-Time Optimization for Dense Correspondence (ICCV 2021 · similarity 0.89)
- Learning Dense Correspondence via 3D-Guided Cycle Consistency (CVPR 2016 · similarity 0.88)
- WarpNet: Weakly Supervised Matching for Single-View Reconstruction (CVPR 2016 · similarity 0.88)
- Discrete-Continuous Transformation Matching for Dense Semantic Correspondence (TPAMI 2018 · similarity 0.88)
- Jamais Vu: Exposing the Generalization Gap in Supervised Semantic Correspondence (NeurIPS 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.
