Paper · ≈ 1 min read
Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors
Outgoing relations
- addresses → Image-to-3D (Task)
- introduces → Magic123 (Algorithm)
- part_of → 3D/4D Vision and Reconstruction (ResearchField)
- uses → Volumetric Rendering (MethodologicalConcept)
- uses_dataset → ShapeNet (Dataset)
- uses_metric → LPIPS (Metric)
- uses_metric → PSNR (Metric)
- uses_metric → SSIM (Metric)
Connected node types
- Metric: 3
- Algorithm: 1
- Dataset: 1
- MethodologicalConcept: 1
- ResearchField: 1
- Task: 1
Mentions in the corpus
- part_of Evidence: Magic123 demonstrates a significant improvement over previous image-to-…
- mentioned_in Claim: Magic123 demonstrates a significant improvement over previous image-to-3D techniques, as valida…
- part_of Evidence: We present Magic123, a two-stage coarse-to-fine approach for high-quali…
- part_of Evidence: We introduce a single trade-off parameter between the 2D and 3D priors…
- part_of Evidence: The second stage swaps the volumetric NeRF for a memory-efficient diffe…
- mentioned_in Claim: The second stage swaps the volumetric NeRF for a memory-efficient differentiable mesh represent…
- part_of Evidence: The two-stage design and the explicit 2D/3D prior interpolation became…
- mentioned_in Claim: The two-stage design and the explicit 2D/3D prior interpolation became common reference points…
- summarizes Project Pulse
- summarizes Field Overview — 3D/4D Vision and Reconstruction
Cross-references in raw data
Source provenance
examples/demo-corpus/data/research/papers/arxiv-2306-17843/abstract.md