Research map / Image generation / Text-to-image diffusion
Enhancing Spatial Understanding in Image Generation via Reward Modeling
Zhenyu Tang, Chaoran Feng, Yufan Deng, Jie Wu, Xiaojie Li, Rui Wang, Yunpeng Chen, Daquan Zhou
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 |
|---|---|---|---|
| scale SpatialScore backbone from Qwen2.5-VL-3B to 7B (ablation) | SpatialScore · Spatial reward benchmark (365 pairs) | pairwise accuracy: 0.891 -> 0.958 | Improved |
| scale SpatialScore backbone from 7B to 32B (ablation) | SpatialScore · Spatial reward benchmark (365 pairs) | pairwise accuracy: 0.958 -> 0.973 | Improved |
| add top-k filtering (k=6) to GRPO (ablation) | Flux.1-dev GRPO · SpatialScore in-domain | SpatialScore reward: 7.73 -> 7.81 | Improved |
| add top-k filtering (k=6) to GRPO (ablation) | Flux.1-dev GRPO · DPG-Bench Relation-Spatial | score: 0.919 -> 0.932 | Improved |
| add top-k filtering (k=6) to GRPO (ablation) | Flux.1-dev GRPO · UniGenBench++ long Layout-2D | score: 0.891 -> 0.887 | Hurt |
| add top-k filtering (k=6) to GRPO (ablation) | Flux.1-dev GRPO · UniGenBench++ long Layout-3D | score: 0.793 -> 0.801 | Improved |
| add top-k filtering (k=6) to GRPO (ablation) | Flux.1-dev GRPO · policy-update NFE per prompt | NFE: 24*6 -> 12*6 | Improved |
| use k=4 instead of k=6 in top-k filtering (ablation) | Flux.1-dev GRPO · SpatialScore in-domain | SpatialScore reward: 7.81 -> 7.71 | Hurt |
| use k=4 instead of k=6 in top-k filtering (ablation) | Flux.1-dev GRPO · DPG-Bench Relation-Spatial | score: 0.932 -> 0.916 | Hurt |
| train Flux.1-dev with SpatialScore RL instead of GenEval reward (Flow-GRPO) | Flux.1-dev · DPG-Bench overall | overall score: 57.02 -> 85.03 | Improved |
| train Flux.1-dev with SpatialScore RL | Flux.1-dev · DPG-Bench overall | overall score: 82.91 -> 85.03 | Improved |
| train Flux.1-dev with SpatialScore RL | Flux.1-dev · SpatialScore in-domain | SpatialScore reward: 2.18 -> 7.81 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- HPSv3 — HPSv3: Towards Wide-Spectrum Human Preference Score (ICCV 2025) · better on 1 of 1 shared result
- ImageReward — ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation (NeurIPS 2023) · better on 1 of 1 shared result
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
- GoT-R1: Unleashing Reasoning Capability of Autoregressive Visual Generation with Reinforcement Learning (ICLR 2026 · similarity 0.89)
- Exploring Spatial Intelligence from a Generative Perspective (CVPR 2026 · similarity 0.88)
- SpatialReward: Bridging the Perception Gap in Online RL for Image Editing via Explicit Spatial Reasoning (ICML 2026 · similarity 0.88)
- Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation (NeurIPS 2025 · similarity 0.87)
- GenSpace: Benchmarking Spatially-Aware Image Generation (NeurIPS 2025 · similarity 0.87)
- Learning to Generate via Understanding: Understanding-Driven Intrinsic Rewarding for Unified Multimodal Models (CVPR 2026 · 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.
