Research map / Image generation / Text-to-image diffusion
TOPA: Mitigating Concept Dominance in Diffusion Personalization via Target-Oriented Perturbation Augmentation
Jiayou Lu, Mingzhi Lyu, Junfeng Huang, Wai-Kin Adams Kong
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 |
|---|---|---|---|
| add TOPA augmentation | ELITE · CustomConcept101 (10 concepts, DreamBench prompts) | FID: 219 -> 176 | Improved |
| add TOPA augmentation | ELITE · CustomConcept101 (10 concepts, DreamBench prompts) | BLIP-VQA accuracy: 0.622 -> 0.681 | Improved |
| add TOPA augmentation | Textual Inversion · CustomConcept101 (10 concepts, DreamBench prompts) | BRISQUE: 16.634 -> 14.574 | Improved |
| add TOPA augmentation | Custom Diffusion · CustomConcept101 (10 concepts, DreamBench prompts) | CLIP-T: 0.506 -> 0.534 | Improved |
| add TOPA augmentation | DreamBooth LoRA · CustomConcept101 (10 concepts, DreamBench prompts) | CLIP-IQA: 0.768 -> 0.908 | Improved |
| add TOPA augmentation | Attend-And-Excite · CustomConcept101 (Attend-And-Excite) | FID: 296 -> 256 | Improved |
| add TOPA augmentation | Pivotal Tuning Inversion LoRA · CustomConcept101 (PTI LoRA) | FID: 187 -> 147 | Improved |
| add naive augmentation instead of TOPA (ablation) | Textual Inversion · CustomConcept101 (10 concepts, DreamBench prompts) | CLIP-I-dagger: 0.667 -> 0.623 | Hurt |
| add naive augmentation instead of TOPA (ablation) | Custom Diffusion · CustomConcept101 (10 concepts, DreamBench prompts) | BRISQUE: 15.147 -> 16.131 | Hurt |
| add background replacement only (no perturbation) (ablation) | Textual Inversion · CustomConcept101 (Textual Inversion ablation) | BLIP-VQA accuracy: 0.577 -> 0.673 | Improved |
| add perturbation optimization on top of background replacement (ablation) | Textual Inversion · CustomConcept101 (Textual Inversion ablation) | FID: 231 -> 220 | Improved |
| add L_ctx loss only (ablation) | Textual Inversion · CustomConcept101 (Textual Inversion ablation) | BRISQUE: 16.634 -> 15.346 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ELITE — ELITE: Encoding Visual Concepts into Textual Embeddings for Customized Text-to-Image Generation (ICCV 2023) · better on 2 of 2 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
- DynASyn: Multi-Subject Personalization Enabling Dynamic Action Synthesis (AAAI 2025 · similarity 0.88)
- How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization? (NeurIPS 2024 · similarity 0.88)
- PaRa: Personalizing Text-to-Image Diffusion via Parameter Rank Reduction (ICLR 2025 · similarity 0.87)
- ClassDiffusion: More Aligned Personalization Tuning with Explicit Class Guidance (ICLR 2025 · similarity 0.87)
- Hollowed Net for On-Device Personalization of Text-to-Image Diffusion Models (NeurIPS 2024 · similarity 0.87)
- APT: Adaptive Personalized Training for Diffusion Models with Limited Data (CVPR 2025 · 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.
