Research map / 3D vision / NeRF and radiance fields
ConRad: Image Constrained Radiance Fields for 3D Generation from a Single Image
Senthil Purushwalkam, Nikhil Naik
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 |
|---|---|---|---|
| use ConRad instead of RealFusion | ConRad · ShapeNet all views | CLIP distance d_oracle (lower better): 0.460 -> 0.273 | Improved |
| use ConRad instead of NeuralLift-360 | ConRad · ShapeNet near reference | CLIP distance d_ref (lower better): 0.543 -> 0.286 | Improved |
| use ConRad instead of PointE | ConRad · ShapeNet all views | CLIP distance d_all (lower better): 0.482 -> 0.316 | Improved |
| remove warm start (ablation) | ConRad · qualitative examples | 3D structure crispness | Mixed |
| remove depth loss (ablation) | ConRad · qualitative examples | 3D structure correctness and compute | Mixed |
| remove textual inversion (ablation) | ConRad · qualitative examples | view-consistent appearance | Hurt |
| remove color constraint (ablation) | ConRad · qualitative examples | realism and fidelity | Hurt |
| remove both color and density constraints, keeping depth loss (ablation) | ConRad · qualitative examples | realism | Hurt |
| use ConRad on one object where baselines win | ConRad · ShapeNet table bdd12e, d_ref | CLIP distance: 0.380 -> 0.560 | Hurt |
| use ConRad on car object 1abeca | ConRad · ShapeNet car 1abeca, d_ref | CLIP distance: 0.458 -> 0.469 | Hurt |
| use textual inversion embedding to mitigate saturated colors from Stable Diffusion (ablation) | ConRad · failure cases | color fidelity | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- RealFusion — RealFusion: 360deg Reconstruction of Any Object From a Single Image (CVPR 2023) · better on 6 of 9 shared results
- NeuralLift-360 — NeuralLift-360: Lifting an In-the-Wild 2D Photo to a 3D Object With 360deg Views (CVPR 2023) · better on 7 of 8 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
- LOLNerf: Learn From One Look (CVPR 2022 · similarity 0.89)
- GRAF: Generative Radiance Fields for 3D-Aware Image Synthesis (NeurIPS 2020 · similarity 0.88)
- Sharf: Shape-conditioned Radiance Fields from a Single View (ICML 2021 · similarity 0.88)
- GRF: Learning a General Radiance Field for 3D Representation and Rendering (ICCV 2021 · similarity 0.87)
- Neural Radiance Fields From Sparse RGB-D Images for High-Quality View Synthesis (TPAMI 2022 · similarity 0.87)
- GaussianCube: A Structured and Explicit Radiance Representation for 3D Generative Modeling (NeurIPS 2024 · 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.
