Research map / 3D vision / Face and shape
Mitigating Perspective Distortion-induced Shape Ambiguity in Image Crops
Aditya Prakash, Arjun Gupta, Saurabh Gupta
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 KPE (sparse) to ArcticNet-SF | ArcticNet-SF · ARCTIC egocentric (test) | Success (%): 53.9 -> 64.1 | Improved |
| use object crops instead of full image (ablation) | ArcticNet-SF · ARCTIC egocentric (validation) | Success (%): 59.0 -> 66.0 | Improved |
| crop input without location info (ablation) | ArcticNet-SF (crop) · ARCTIC egocentric (validation) | CDev (mm): 44.1 -> 74.3 | Hurt |
| encode location in camera field of view with sinusoids (KPE) instead of 2D image location (ablation) | ArcticNet-SF (crop) · ARCTIC egocentric (validation) | Success (%): 67.9 -> 71.5 | Improved |
| encode both scale and location (corners and center) instead of center only (ablation) | sparse KPE · ARCTIC egocentric (validation) | Success (%): 66.8 -> 71.5 | Improved |
| feed dense KPE to the 7x7 feature map instead of the input (ablation) | dense KPE · ARCTIC egocentric (validation) | Success (%): 68.1 -> 71.4 | Improved |
| add dense KPE to ZoeDepth (no pretraining) | ZoeD-X-N · NYU Depth v2 | REL: 0.093 -> 0.088 | Improved |
| add dense KPE to ZoeDepth (pretrained MiDaS) | ZoeD-M12-N · NYU Depth v2 | REL: 0.088 -> 0.082 | Improved |
| add Cam-Conv to ZoeDepth | ZoeD-M12-N · NYU Depth v2 | REL: 0.088 -> 0.096 | Hurt |
| inject KPE inside the MiDaS backbone before BeiT instead of after (ablation) | ZoeD-M12-N + KPE (dense) · NYU Depth v2 | REL: 0.085 -> 0.082 | Improved |
| add KPE (sparse) to Cube R-CNN | Cube R-CNN · KITTI + nuScenes Cars (joint) | AP3D: 54.59 -> 57.54 | Improved |
| use sparse KPE instead of dense KPE for detection (ablation) | Cube R-CNN + KPE · KITTI + nuScenes Cars (joint) | AP3D: 56.78 -> 57.54 | Improved |
Closest papers by meaning
- PCLs: Geometry-Aware Neural Reconstruction of 3D Pose With Perspective Crop Layers (CVPR 2021 · similarity 0.87)
- Know Your Neighbors: Improving Single-View Reconstruction via Spatial Vision-Language Reasoning (CVPR 2024 · similarity 0.85)
- Cameras as Relative Positional Encoding (NeurIPS 2025 · similarity 0.85)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.85)
- RefineAny3D: Depth Refinement as Semantic Alignment for Monocular 3D Detection (NeurIPS 2026 · similarity 0.85)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.84)
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.
