Although debiased large language models (LLMs) excel at handling known or low-bias prompts, they often fail on unfamiliar and high-bias prompts.We demonstrate via out-ofdistribution (OOD) detection that these highbias prompts cause a distribution shift, degrading static model performance.To enable realtime correction, we propose CAP-TTA, a testtime adaptation framework.CAP-TTA triggers context-aware LoRA updates only when a bias-risk score exceeds a set threshold.By utilizing an offline precomputed diagonal preconditioner, it ensures fast and stable optimization.Across multiple benchmarks and human evaluations, CAP-TTA effectively reduces toxicity/bias score with significantly lower latency than standard optimization methods (e.g., AdamW or SGD).Furthermore, it prevents catastrophic forgetting, and substantially improves narrative fluency over state-of-the-art baselines without compromising debiasing performance.

The rapid progress of generative AI has led to the emergence of new generative models, while existing detection methods struggle to keep pace with new model series and architectures, resulting in significant degradation in the detection performance. This highlights the urgent need for continuously updating AI-generated image detectors to adapt to new generators. To overcome low efficiency and catastrophic forgetting in detector updates, we propose LiteUpdate, a lightweight framework for updating AI-generated image detectors to unseen generative models. Unlike previous approaches that use randomly sampled training data, LiteUpdate employs a representative sample selection module that leverages image confidence and gradient-based discriminative features to precisely select boundary samples. This approach improves learning and detection accuracy on new distributions with limited generated images, significantly enhancing detector update efficiency. Additionally, LiteUpdate incorporates a model merging module that fuses weights from multiple fine-tuning trajectories, including pre-trained, representative, and random updates. This balances the adaptability to new generators and mitigates the catastrophic forgetting of previously learned knowledge. Experiments demonstrate that LiteUpdate substantially boosts detection performance in various detectors with high efficiency. Specifically, on AIDE, the average detection accuracy on Midjourney improved from 87.63% to 93.03%, a 6.16% relative increase. Meanwhile, to achieve comparable accuracy, LiteUpdate attains approximately 4× speedup over conventional random sample fine-tuning.

In large vision-language models (LVLMs), images serve as inputs that carry a wealth of information. As the idiom "A picture is worth a thousand words" implies, representing a single image in current LVLMs can require hundreds or even thousands of tokens, which thereby severely impacting the efficiency. Previous approaches have attempted to reduce the number of image tokens either before or within the early layers of LVLMs. However, these strategies inevitably result in the loss of crucial image information. To address this challenge, we conduct an empirical study revealing that all visual tokens are necessary for LVLMs in the shallow layers, and token redundancy progressively increases in the deeper layers. To this end, we propose PyramidDrop, a visual redundancy reduction strategy for LVLMs to boost their efficiency in both training and inference with neglectable performance loss. Specifically, we partition the LVLM into several stages and drop part of the image tokens at the end of each stage with a pre-defined ratio. The dropping is based on a lightweight similarity calculation with a negligible time overhead. Extensive experiments demonstrate that PyramidDrop can achieve over 40% training time reduction and 55% inference FLOPs acceleration on leading LVLMs like LLaVA-NeXT, maintaining comparable multi-modal performance. Besides, PyramidDrop can also serve as a plug-and-play strategy to accelerate inference in a free way, with better performance and lower inference cost than counterparts. Our code is available at https://github.com/Cooperx521/PyramidDrop.

Purely rotation-based self-calibration receives the most attention among various self-calibration methods owing to its algorithmic simplicity. However, it is actually impossible to ensure that the camera motion for this kind of self-calibration is a pure rotation. Thus, significant errors of calibrating results could be inevitably introduced because of ignoring nonzero translation. In this paper, we propose a practical and effective approach to improve the purely rotation-based self-calibration approach. According to the fact that the rotational angles between images have a very strong impact on the calibrating errors from the translations, we compute the rotational angles between images prior to calibrating, and then use different and very appropriate strategies for self-calibration in different angle circumstances. Real data has been used to validate the proposed approach.