Research map / Image generation / Text-to-image diffusion
CLIP-Forge: Towards Zero-Shot Text-To-Shape Generation
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 |
|---|---|---|---|
| Generate shapes with CLIP-Forge without paired text-shape labels rather than using the supervised text2shape baselines | CLIP-Forge · Text query set (ShapeNet) | FID: 14881.96 -> 2425.25 | Helped |
| Add Gaussian noise to the shape latent (ablation) | CLIP-Forge autoencoder · Text query set (ShapeNet) | FID: 3871.48 -> 2688.72 | Helped |
| Increase latent size from 128 to 256 or 512 (ablation) | CLIP-Forge autoencoder · Text query set (ShapeNet) | FID: 2688.72 -> 3177.92; 2688.72 -> 3577.72 | Hurt |
| Replace VoxEnc with PointNet (ablation) | CLIP-Forge autoencoder · Text query set (ShapeNet) | FID: 2688.72 -> 2646.93 | Helped |
| Replace VoxEnc with ResVoxEnc (ablation) | CLIP-Forge autoencoder · Text query set (ShapeNet) | FID: 2688.72 -> 3146.94 | Hurt |
| Replace RN-OccNet with CBN-OccNet (ablation) | CLIP-Forge autoencoder · Text query set (ShapeNet) | FID: 2688.72 -> 2956.78 | Hurt |
| Condition affine coupling layers rather than conditioning only the prior network (ablation) | RealNVP-C · Text query set (ShapeNet) | FID: 5227.32 -> 2688.72 | Helped |
| Use dimension-wise rather than checkerboard masking (ablation) | RealNVP · Text query set (ShapeNet) | FID: 2688.72 -> 2591.87 | Helped |
| Replace RealNVP with MAF (ablation) | CLIP-Forge flow prior · Text query set (ShapeNet) | FID: 2591.87 -> 6052.62 | Hurt |
| Use additional shape renderings to train the conditional prior (ablation) | CLIP-Forge · Text query set (ShapeNet) | FID: 2983.72 -> 2776.28 -> 2622.71 -> 2591.87 | Helped |
| Replace CLIP ViT-B/32 with ViT-B/16 (ablation) | CLIP-Forge · Text query set (ShapeNet) | FID: 2591.87 -> 2515.81 | Helped |
| Replace CLIP ViT-B/32 with RN50x16 (ablation) | CLIP-Forge · Text query set (ShapeNet) | FID: 2591.87 -> 2906.75 | Hurt |
Closest papers by meaning
- ShapeWords: Guiding Text-to-Image Synthesis with 3D Shape-Aware Prompts (CVPR 2025 · similarity 0.87)
- CLIP in Mirror: Disentangling text from visual images through reflection (NeurIPS 2024 · similarity 0.86)
- Seeing What Matters: Empowering CLIP with Patch Generation-to-Selection (CVPR 2025 · similarity 0.86)
- LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 2024 · similarity 0.85)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.85)
- ShapeWalk: Compositional Shape Editing Through Language-Guided Chains (CVPR 2024 · similarity 0.85)
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.