30 January 2025 in IEEE Transactions on Pattern Analysis and Machine Intelligence
Recent source-free domain adaptation (SFDA) methods have focused on learning meaningful cluster structures in feature space, successfully adapting the knowledge from the source domain to the unlabeled target domain without accessing the private source data. However, existing methods rely on pseudo-labels generated by source models that can be noisy due to domain shift, presenting a significant challenge to their efficacy. In this paper, we study SFDA from the perspective of learning with label noise (LLN) and prove that the label noise in SFDA, unlike in conventional LLN scenarios, follows a different distribution assumption. This discrepancy renders some existing LLN methods less effective in SFDA. To address this issue and comprehensively improve adaptation performance, we tackle label noise in SFDA from two perspectives. First, we demonstrate that the early-time training phenomenon (ETP), previously observed in LLN settings, still exists in SFDA. Hence, we introduce a simple yet effective approach to leveraging ETP to improve current SFDA algorithms. Second, we propose a noise and variance control module, mitigating the label noise discrepancy between SFDA and LLN and enhancing the effectiveness of LLN methods in SFDA. Extensive empirical evaluation and analysis of four benchmarks show that our methods substantially outperform existing baselines.
19 August 2025 in TechRxiv
The centrality of image editing and controllable generation lies in conditioning desired visual content, which maintains the consistency and invariance of certain parts of the image. Leveraging the large-scale flow transformer for these tasks requires two key aspects: the authentic inversion that projects the image into the model's domain and the flexible invariance control mechanism to preserve non-target contents. However, the prevailing diffusion inversion performs deficiently in flow-based models, and the invariance control cannot reconcile diverse rigid and non-rigid editing tasks. To address these, we theoretically analyze the inversion and invariance control based on the flow transformer. Specifically, we unveil that the Euler inversion shares a similar structure to DDIM yet is more susceptible to the approximation error. Thus, we propose a two-stage inversion to first refine the velocity estimation and then compensate for the leftover error, which pivots closely to the model prior and benefits editing. We also provide the convergence analysis of the proposed inversion for rectified flow. Meanwhile, we propose the invariance control that manipulates the text features within the adaptive layer normalization, connecting the changes in the text prompt to image semantics. This mechanism can simultaneously preserve the non-target contents while allowing both rigid and non-rigid manipulation, enabling a wide range of editing types such as visual text, quantity, facial expression, etc. Furthermore, we show that this mechanism can unify the invariance control in both image editing and controllable generation, such as ID consistency. Experiments on versatile scenarios of image editing and ID consistent generation validate that our approach achieves flexible editing and controllable generation within a unified framework, unlocking the potential of the flow transformer for the general conditional image generation tasks.