Research map / 3D vision / NeRF and radiance fields
Pix2NeRF: Unsupervised Conditional p-GAN for Single Image to Neural Radiance Fields Translation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Condition on an input image rather than sample content codes from the prior | Pix2NeRF · CelebA 64 × 64 | FID: 6.25 -> 24.64 | Hurt |
| Condition on an input image rather than sample content codes from the prior | Pix2NeRF · CARLA 64 × 64 | FID: 10.54 -> 12.06 | Hurt |
| Add Pix2NeRF training to the π-GAN backbone | CARLA · CARLA 64 × 64 | FID: 13.59 -> 10.54 | Helped |
| Add Pix2NeRF training to the π-GAN backbone | ShapeNet-SRN chairs · ShapeNet-SRN chairs 64 × 64 | FID: 15.47 -> 14.31 | Helped |
| Freeze the pretrained generator and train only an encoder (ablation) | Pix2NeRF · CelebA 64 × 64 | FID: 24.64 -> 28.90 | Hurt |
| Drop the latent-space GAN path and train only reconstruction and conditional adversarial objectives (ablation) | Pix2NeRF · CelebA 64 × 64 | FID: 24.64 -> 43.19 | Hurt |
| Remove the GAN inversion objective (ablation) | Pix2NeRF · CelebA 64 × 64 | FID: 24.64 -> 39.42 | Hurt |
| Remove the conditional adversarial objective (ablation) | Pix2NeRF · CelebA 64 × 64 | FID: 24.64 -> 33.92 | Hurt |
| Train the encoder from the start instead of freezing it during warm-up (ablation) | Pix2NeRF · CelebA 64 × 64 | FID: 24.64 -> 31.31 | Hurt |
| Keep the encoder frozen throughout training (ablation) | Pix2NeRF · CelebA 64 × 64 | FID: 24.64 -> 39.86 | Hurt |
| Replace warm-up with a lower reconstruction weight λ_recon=1 (ablation) | Pix2NeRF · CelebA 64 × 64 | FID: 24.64 -> 73.52 | Hurt |
| Replace warm-up with a lower reconstruction weight λ_recon=0.1 (ablation) | Pix2NeRF · CelebA 64 × 64 | FID: 24.64 -> 73.03 | Hurt |
Closest papers by meaning
- GRAF: Generative Radiance Fields for 3D-Aware Image Synthesis (NeurIPS 2020 · similarity 0.91)
- Shape, Pose, and Appearance From a Single Image via Bootstrapped Radiance Field Inversion (CVPR 2023 · similarity 0.90)
- Benchmarking and Analyzing 3D-aware Image Synthesis with a Modularized Codebase (NeurIPS 2023 · similarity 0.90)
- pixelNeRF: Neural Radiance Fields From One or Few Images (CVPR 2021 · similarity 0.90)
- In-N-Out: Faithful 3D GAN Inversion with Volumetric Decomposition for Face Editing (CVPR 2024 · similarity 0.87)
- Mask-Based Modeling for Neural Radiance Fields (ICLR 2024 · 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.