Research map / 3D vision / Face and shape
3DMiner: Discovering Shapes from Large-Scale Unannotated Image Datasets
Ta-Ying Cheng, Matheus Gadelha, Sören Pirk, Thibault Groueix, Radomír Měch, Andrew Markham, Niki Trigoni
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 augmentations (color jitter, rotation, perspective) with pooled DINO-ViT features before clustering (ablation) | 3DMiner clustering · Pix3D | NMI: 0.647 -> 0.758 | Improved |
| vary agglomerative clustering distance threshold (10, 30, 100) (ablation) | 3DMiner clustering with augmentation · Pix3D | NMI: 0.758 / 0.786 / 0.764 | No clear effect |
| filter reconstructions by lower per-cluster reprojection error | 3DMiner · Pix3D Chairs | F1: 0.234 -> 0.307 | Improved |
| replace SMR/Unicorn with 3DMiner pipeline | unsupervised 3D reconstruction · Pix3D Chairs | Chamfer Distance: 0.192 -> 0.130 (vs SMR) | Improved |
| bundle-adjust motion and translation plus focal length, versus no or partial adjustment (ablation) | 3DMiner occupancy field · single accurate Pix3D chair cluster | qualitative reconstruction | Improved |
| train Unicorn on real Pix3D chairs | Unicorn baseline · Pix3D Chairs | reconstruction | Hurt |
| tighten clustering threshold to reduce non-rigidity | 3DMiner · LAION-5B | reconstruction success | Mixed |
Closest papers by meaning
- ShapeClipper: Scalable 3D Shape Learning From Single-View Images via Geometric and CLIP-Based Consistency (CVPR 2023 · similarity 0.89)
- Shape and Viewpoint without Keypoints (ECCV 2020 · similarity 0.88)
- MESC-3D:Mining Effective Semantic Cues for 3D Reconstruction from a Single Image (CVPR 2025 · similarity 0.88)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 2025 · similarity 0.87)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.87)
- Stereo4D: Learning How Things Move in 3D from Internet Stereo Videos (CVPR 2025 · similarity 0.87)
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.
