无需重训练，按需调节文本生成图像的性别偏向。
EquiSteer: Cross-Attention Steering Towards a Fairer Text-Guided Image Generation
- 推理时动态调整跨注意力激活，实现细粒度性别控制。
- 在多个模型上将性别偏差降低87%，保持图像质量与文本对齐。
- 适合需要公平生成、无需调参的创意应用者使用。
文生图扩散模型虽广泛用于日常创作，但仍会继承训练数据中的种族与性别偏见。例如在生成‘护士’或‘首席执行官’等提示词时，模型输出倾向某一性别，这是由训练数据统计驱动，而非文本本身。现有去偏方法多需重新训练、批量控制或针对特定提示微调，难以扩展。本文提出EquiSteer，一种无需训练的方法，在生成阶段通过推理时调整跨注意力（CA）激活来实现性别均衡。针对每个目标属性，预先从对比提示中计算引导向量；生成时，对特定属性提示保留原信号，对中性提示则清除原有属性信息并注入目标属性。在SD-1.5、SD-2.1、SDXL和SANA模型上，平均性别偏差减少高达87%，且对图像质量与图文一致性影响极小。代码已开源。
原文摘要 · Abstract (English)
Text-to-image diffusion models power everyday creative tasks, but they still reproduce the demographic biases in their training data. On common prompts such as ``a photo of a nurse,'' ``a photo of a CEO'', they skew their outputs toward one gender, driven by the statistics of training data rather than anything in the text. Existing debiasing methods show promise in narrow settings but require retraining, batch-level control, or prompt-specific tuning, limiting their scalability. We propose \emph{EquiSteer}, a training-free method that works per sample by steering cross-attention (CA) activations at inference time. For each target attribute, EquiSteer precomputes steering vectors from contrastive prompts. Then at generation time, a prompt-aware gate leaves attribute-specific prompts untouched, while for neutral ones it clears existing attribute signals from the CA activations and injects a target attribute. Across SD-1.5, SD-2.1, SDXL, and SANA, EquiSteer reduces the average parity gap by up to $87\%$, with minimal effect on image quality and text-image alignment. Code is available at \href{https://github.com/Atmyre/EquiSteer}{https://github.com/Atmyre/EquiSteer}.%
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。