Fine-grained species classification involves distinguishing visually similar taxa and remains challenging for general-purpose visual recognition models. We propose BioReason, a retrieve-then-reasonframework that uses BioCLIP candidates to formulate a multiple-choice task for Qwen2.5-VL-7B, followed by two-source superv...
The rapid deployment of Text-to-Image (T2I) diffusion models has raised urgent concerns about the generation of unsafe, copyrighted, and offensive content. Recently, steering-based concept unlearning methods have emerged as an efficient alternative to retraining, suppressing undesired concepts by injecting learned dire...
Yan-Bin Hao, Chen He, Shuo Wang et al.· 0 citations
Representation alignment (REPA) is widely used to accelerate the training of latent-space diffusion transformers (DiTs). However, applying it to pixel-space models such as Just Image Transformers (JiT) is challenging. Our analysis reveals a pronounced representation conflict between discriminative alignment and pixel-s...
Yi Tan, Bei-Er Zhu, Si-Rui Chen et al.· 0 citations
Across continual erasure of celebrities, artistic styles, and instances, CEASE achieves the most consistent erase-preserve trade-off, while existing methods either degrade general generation or insufficiently erase targets.
Yong-Liang Wu, Haori Lu, Jin-Qi Luo et al.· 0 citations
Retain-aware Activation Steering (RASteer), a training-free method that first builds a retain subspace from the concepts to preserve, and introduces Overlap-Adaptive Calibration (OAC), which uses the overlap between the erasure direction and the retain subspace to control how much of each shared component is removed, b...
Yong-Liang Wu, Haori Lu, Yu-Lun Wu et al.· 1 citation
Text-to-motion generation and motion-to-text captioning are two fundamental tasks in human motion modeling, both grounded in the same underlying motion-text correspondence. Existing unified approaches mostly rely on autoregressive modeling, which imposes a fixed generation order and is therefore poorly suited to the bi...
Wan-Jiang Weng, Yong-Liang Wu, Xiaofeng Tan et al.· 0 citations
Perception-Enhanced Alignment DPO (PEA-DPO), a framework for multimodal LLMs alignment, which explicitly leverages visual preference signals to overcome visual insensitivity is proposed, which demonstrates that PEA-DPO enhances sensitivity to visual context while preserving the language modeling capacity of the base mo...
Jiawei Feng, Jiancan Wu, Xingyu Zhu et al.· 1 citation
This work proposes RITA, a Robust test-tIme prompt-TAdaptation framework that shifts from sample-level estimates to distribution-level alignment, and employs optimal transport to align the distribution of augmented visual features with textual prototypes, mitigating adversarial outliers and rectifying cross-modal seman...
Xingyu Zhu, Huanshen Wu, Shuo Wang et al.· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.