Research map / 3D vision / Depth and stereo
Content-Aware Unsupervised Deep Homography Estimation
Jirong Zhang, Chuan Wang, Shuaicheng Liu, Lanpeng Jia, Nianjin Ye, Jue Wang, Ji Zhou, Jian Sun
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 content-aware mask (attention and loss weighting) (ablation) | Content-aware homography network · Homography dataset, all scenes avg | mean l2 error (pixels): 2.38 -> 1.82 | Improved |
| use mask as attention map and loss weight rather than attention only (ablation) | Content-aware homography network · Homography dataset, all scenes avg | mean l2 error (pixels): 2.27 -> 1.82 | Improved |
| use mask as attention map in addition to loss weighting (ablation) | Content-aware homography network · Homography dataset, all scenes avg | mean l2 error (pixels): 2.07 -> 1.82 | Improved |
| add triplet loss term (ablation) | Content-aware homography network · Homography dataset, all scenes avg | mean l2 error (pixels): 2.84 -> 1.82 | Improved |
| add feature extractor instead of raw pixel intensity (ablation) | Content-aware homography network · Homography dataset, all scenes avg | mean l2 error (pixels): 2.03 -> 1.82 | Improved |
| use ResNet-34 instead of VGG backbone (ablation) | Content-aware homography network · Homography dataset, all scenes avg | mean l2 error (pixels): 2.17 -> 1.82 | Improved |
| use ResNet-34 instead of ResNet-18 backbone (ablation) | Content-aware homography network · Homography dataset, all scenes avg | mean l2 error (pixels): 2.06 -> 1.82 | No clear effect |
| use ResNet-34 instead of ShuffleNet-v2 backbone (ablation) | Content-aware homography network · Homography dataset, all scenes avg | mean l2 error (pixels): 2.44 -> 1.82 | Improved |
| add two-stage training (attention disabled first, then enabled) (ablation) | Content-aware homography network · Homography dataset, all scenes avg | mean l2 error (pixels): 1.90 -> 1.82 | Improved |
| use proposed method instead of SIFT + RANSAC | Content-aware homography network · Homography dataset, LT low texture scenes | mean l2 error (pixels): 2.56 -> 1.90 | Improved |
| use proposed method instead of SIFT + MAGSAC | Content-aware homography network · Homography dataset, RE regular scenes | mean l2 error (pixels): 1.71 -> 1.81 | Hurt |
| use proposed method instead of LIFT + MAGSAC | Content-aware homography network · Homography dataset, all scenes avg | mean l2 error (pixels): 2.07 -> 1.82 | Improved |
Closest papers by meaning
- Adapting Dense Matching for Homography Estimation with Grid-based Acceleration (CVPR 2025 · similarity 0.90)
- Mask-Homo: Pseudo Plane Mask-Guided Unsupervised Multi-Homography Estimation (AAAI 2024 · similarity 0.89)
- Unsupervised Homography Estimation on Multimodal Image Pair via Alternating Optimization (NeurIPS 2024 · similarity 0.89)
- MCNet: Rethinking the Core Ingredients for Accurate and Efficient Homography Estimation (CVPR 2024 · similarity 0.89)
- SSHNet: Unsupervised Cross-modal Homography Estimation via Problem Reformulation and Split Optimization (CVPR 2025 · similarity 0.89)
- Noisy One-point Homographies are Surprisingly Good (CVPR 2024 · 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.
