Research map / Image generation / Text-to-image diffusion
Towards Reliable Advertising Image Generation Using Human Feedback
Zhenbang Du, Wei Feng, Haohan Wang, Yaoyu Li, Jingsen Wang, Jian Li, Zheng Zhang, Jingjing Lv, Xin Zhu, Junsheng Jin, Junjie Shen, Zhangang Lin and 1 more
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 product image I_o as RFNet input (ablation) | RFNet · RF1M inspection | AP: 81.17 -> 87.58 | Improved |
| add generated image I_g (as separate input) (ablation) | RFNet · RF1M inspection | AP: 82.06 -> 87.58 | Improved |
| add depth image I_d (ablation) | RFNet · RF1M inspection | AP: 85.31 -> 87.58 | Improved |
| add salience image I_s (ablation) | RFNet · RF1M inspection | AP: 83.91 -> 87.58 | Improved |
| add product caption with Feature Filter Module (ablation) | RFNet · RF1M inspection | AP: 84.53 -> 87.58 | Improved |
| use five fine-grained labels instead of coarse Available/Unavailable (ablation) | RFNet · RF1M inspection | AP: 82.06 -> 87.58 | Improved |
| replace single-modality baselines with multi-modal RFNet | inspection models · RF1M inspection | AP: 79.62 -> 87.58 (vs ResNeXt50) | Improved |
| back-propagate RFNet feedback directly without regularization (ablation) | diffusion model fine-tuning · image quality | aesthetics / background diversity: 99.8% available rate with collapsed background | Hurt |
| increase beta for KL regularization (ablation) | RFFT · available rate | available rate (%) | Hurt |
| replace KL regularization with Consistent Condition regularization (ablation) | RFFT · available rate vs beta | available rate (%) | Improved |
| add ImageReward feedback F_IR with F_AC (no CC) (ablation) | RFFT · generated image quality | background quality | Mixed |
| add ImageReward feedback F_IR on top of F_AC + L_CC (ablation) | RFFT · background aesthetics | aesthetics | No clear effect |
Closest papers by meaning
- RefAdGen: High-Fidelity Advertising Image Generation (AAAI 2026 · similarity 0.89)
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · similarity 0.86)
- GenAI Arena: An Open Evaluation Platform for Generative Models (NeurIPS 2024 · similarity 0.84)
- ImagenHub: Standardizing the evaluation of conditional image generation models (ICLR 2024 · similarity 0.84)
- Image Regeneration: Evaluating Text-to-Image Model via Generating Identical Image with Multimodal Large Language Models (AAAI 2025 · similarity 0.84)
- OneActor: Consistent Subject Generation via Cluster-Conditioned Guidance (NeurIPS 2024 · similarity 0.83)
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.
