Research map / Image generation / Text-to-image diffusion
Zero-Shot Depth Aware Image Editing with Diffusion Models
Rishubh Parihar, Sachidanand VS, R. Venkatesh Babu
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 FeatGLaC guided insertion instead of Anydoor | depth-aware object insertion · Depth Edit Benchmark: object insertion (490 pairs) | CLIP-sim: 83.23 -> 84.86 | Improved |
| use FeatGLaC guided insertion instead of Anydoor | depth-aware object insertion · Depth Edit Benchmark: object insertion (490 pairs) | DINO-sim: 0.507 -> 0.545 | Improved |
| use FeatGLaC guided insertion instead of Anydoor | depth-aware object insertion · Depth Edit Benchmark: object insertion (490 pairs) | Delta depth: 3.176 -> 2.989 | Improved |
| use FeatGLaC guided insertion instead of PbE | depth-aware object insertion · Depth Edit Benchmark: object insertion (490 pairs) | CLIP-sim: 60.12 -> 84.86 | Improved |
| use DeGLaD Diff composition instead of DeGLaD Image | scene composition · Depth Edit Benchmark: scene composition (2,844 pairs) | FID: 132.6 -> 123.32 | Improved |
| use DeGLaD Diff composition instead of DeGLaD Image | scene composition · Depth Edit Benchmark: scene composition (2,844 pairs) | LPIPS: 0.036 -> 0.263 | Hurt |
| use DeGLaD Diff composition instead of Pair-Diffusion | scene composition · Depth Edit Benchmark: scene composition (2,844 pairs) | FID: 140.54 -> 123.32 | Improved |
| guide only for early timesteps (0-20) instead of 0-38 (ablation) | FeatGLaC scene composition · scene composition (qualitative) | structure preservation | Hurt |
| guide for all timesteps (0-50) instead of 0-38 (ablation) | FeatGLaC scene composition · scene composition (qualitative) | realism | Hurt |
| guide with all decoder layers instead of layers 2 and 3 (ablation) | FeatGLaC scene composition · scene composition (qualitative) | artifacts | Hurt |
| guide with only one decoder layer instead of both layers 2 and 3 (ablation) | FeatGLaC scene composition · scene composition (qualitative) | composition quality | Hurt |
| use foreground guidance weight lambda=1 instead of a smaller value (ablation) | FeatGLaC scene composition · scene composition (qualitative) | composition quality: lambda = 0.25 -> lambda = 1 | Improved |
Closest papers by meaning
- Move Anything with Layered Scene Diffusion (CVPR 2024 · similarity 0.89)
- Stable Flow: Vital Layers for Training-Free Image Editing (CVPR 2025 · similarity 0.89)
- PAIR Diffusion: A Comprehensive Multimodal Object-Level Image Editor (CVPR 2024 · similarity 0.88)
- Generative Image Layer Decomposition with Visual Effects (CVPR 2025 · similarity 0.88)
- Early Semantic Grounding in Image Editing Models for Zero-Shot Referring Image Segmentation (NeurIPS 2026 · similarity 0.88)
- Diffusion Handles Enabling 3D Edits for Diffusion Models by Lifting Activations to 3D (CVPR 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.
