Beyond the Dirac Delta: Mitigating Diversity Collapse in Reinforcement Fine-Tuning for Versatile Image Generation

• URL: http://arxiv.org/abs/2601.12401v1
• Date: Sun, 18 Jan 2026 13:25:43 GMT
• Title: Beyond the Dirac Delta: Mitigating Diversity Collapse in Reinforcement Fine-Tuning for Versatile Image Generation
• Authors: Jinmei Liu, Haoru Li, Zhenhong Sun, Chaofeng Chen, Yatao Bian, Bo Wang, Daoyi Dong, Chunlin Chen, Zhi Wang,
• Abstract summary: We propose textbfDRIFT (textbfDivetextbfRsity-textbfIncentivized Reinforcement textbfFine-textbfTuning for Versatile Image Generation), an innovative framework that systematically incentivizes output throughout the on-policy fine-tuning process.<n>DRIFT achieves superior dominance regarding task alignment and generation diversity, yielding a $ 9.08%!sim! 43.46%$ increase in diversity equivalent alignment levels and a $ 59.65
• Score: 51.305316234962554
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Reinforcement learning (RL) has emerged as a powerful paradigm for fine-tuning large-scale generative models, such as diffusion and flow models, to align with complex human preferences and user-specified tasks. A fundamental limitation remains \textit{the curse of diversity collapse}, where the objective formulation and optimization landscape inherently collapse the policy to a Dirac delta distribution. To address this challenge, we propose \textbf{DRIFT} (\textbf{D}ive\textbf{R}sity-\textbf{I}ncentivized Reinforcement \textbf{F}ine-\textbf{T}uning for Versatile Image Generation), an innovative framework that systematically incentivizes output diversity throughout the on-policy fine-tuning process, reconciling strong task alignment with high generation diversity to enhance versatility essential for applications that demand diverse candidate generations. We approach the problem across three representative perspectives: i) \textbf{sampling} a reward-concentrated subset that filters out reward outliers to prevent premature collapse; ii) \textbf{prompting} with stochastic variations to expand the conditioning space, and iii) \textbf{optimization} of the intra-group diversity with a potential-based reward shaping mechanism. Experimental results show that DRIFT achieves superior Pareto dominance regarding task alignment and generation diversity, yielding a $ 9.08\%\!\sim\! 43.46\%$ increase in diversity at equivalent alignment levels and a $ 59.65\% \!\sim\! 65.86\%$ increase in alignment at equivalent levels of diversity.

Related papers

• DiverseGRPO: Mitigating Mode Collapse in Image Generation via Diversity-Aware GRPO [50.89703227426486]
  Reinforcement learning (RL) improves image generation quality significantly by comparing the relative performance of images generated within the same group.<n>In the later stages of training, the model tends to produce homogenized outputs, lacking creativity and visual diversity.<n>This issue can be analyzed from both reward modeling and generation dynamics perspectives.
  arXiv  Detail & Related papers  (2025-12-25T05:37:37Z)
• DiverseAR: Boosting Diversity in Bitwise Autoregressive Image Generation [22.400053095939402]
  We introduce DiverseAR, a principled and effective method that enhances image diversity without sacrificing visual quality.<n>Specifically, we introduce an adaptive logits distribution scaling mechanism that dynamically adjusts the sharpness of the binary output distribution during sampling.<n>To mitigate potential fidelity loss caused by distribution smoothing, we develop an energy-based generation path search algorithm that avoids sampling low-confidence tokens.
  arXiv  Detail & Related papers  (2025-12-02T16:54:36Z)
• Training-Free Generation of Diverse and High-Fidelity Images via Prompt Semantic Space Optimization [50.5332987313297]
  We propose Token-Prompt embedding Space Optimization (TPSO), a training-free and model-agnostic module.<n>TPSO introduces learnable parameters to explore underrepresented regions of the token embedding space, reducing the tendency of the model to repeatedly generate samples from strong modes of the learned distribution.<n>In experiments on MS-COCO and three diffusion backbones, TPSO significantly enhances generative diversity, improving baseline performance from 1.10 to 4.18 points, without sacrificing image quality.
  arXiv  Detail & Related papers  (2025-11-25T00:42:09Z)
• DisCo: Reinforcement with Diversity Constraints for Multi-Human Generation [60.741022906593685]
  DisCo is the first RL-based framework to directly optimize identity diversity in multi-human generation.<n>DisCo fine-tunes flow-matching models via Group-Relative Policy Optimization.<n>On the DiverseHumans Testset, DisCo achieves 98.6 Unique Face Accuracy and near-perfect Global Identity Spread.
  arXiv  Detail & Related papers  (2025-10-01T19:28:51Z)
• Bayesian-Guided Diversity in Sequential Sampling for Recommender Systems [1.675857332621569]
  We propose a novel framework that leverages a multi-objective, contextual sequential sampling strategy.<n>Item selection is guided by Bayesian updates that dynamically adjust scores to optimize diversity.<n> Experiments on a real-world dataset show that our approach significantly improves diversity without sacrificing relevance.
  arXiv  Detail & Related papers  (2025-06-22T19:36:02Z)
• Evaluating the Diversity and Quality of LLM Generated Content [72.84945252821908]
  We introduce a framework for measuring effective semantic diversity--diversity among outputs that meet quality thresholds.<n>Although preference-tuned models exhibit reduced lexical and syntactic diversity, they produce greater effective semantic diversity than SFT or base models.<n>These findings have important implications for applications that require diverse yet high-quality outputs.
  arXiv  Detail & Related papers  (2025-04-16T23:02:23Z)
• Few-Shot Image Generation by Conditional Relaxing Diffusion Inversion [37.18537753482751]
  Conditional Diffusion Relaxing Inversion (CRDI) is designed to enhance distribution diversity in synthetic image generation. CRDI does not rely on fine-tuning based on only a few samples. It focuses on reconstructing each target image instance and expanding diversity through few-shot learning.
  arXiv  Detail & Related papers  (2024-07-09T21:58:26Z)
• Source-free Domain Adaptation Requires Penalized Diversity [60.04618512479438]
  Source-free domain adaptation (SFDA) was introduced to address knowledge transfer between different domains in the absence of source data. In unsupervised SFDA, the diversity is limited to learning a single hypothesis on the source or learning multiple hypotheses with a shared feature extractor. We propose a novel unsupervised SFDA algorithm that promotes representational diversity through the use of separate feature extractors.
  arXiv  Detail & Related papers  (2023-04-06T00:20:19Z)
• Auto-regressive Image Synthesis with Integrated Quantization [55.51231796778219]
  This paper presents a versatile framework for conditional image generation. It incorporates the inductive bias of CNNs and powerful sequence modeling of auto-regression. Our method achieves superior diverse image generation performance as compared with the state-of-the-art.
  arXiv  Detail & Related papers  (2022-07-21T22:19:17Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.