Benefiting from the advances in large language models and cross-modal alignment, existing multimodal large language models have achieved prominent performance in image and short video understanding. However, the understanding of long videos is still challenging, as their long-context nature results in significant computational and memory overhead. Most existing work treats long videos in the same way as short videos, which is inefficient for real-world applications and hard to generalize to even longer videos. To address these issues, we propose Flash-VStream, an efficient video language model capable of processing extremely long videos and responding to user queries in real time. Particularly, we design a Flash Memory module, containing a low-capacity context memory to aggregate long-context temporal information and model the distribution of information density, and a high-capacity augmentation memory to retrieve detailed spatial information based on this distribution. Compared to existing models, Flash-VStream achieves significant reductions in inference latency. Extensive experiments on long video benchmarks and comprehensive video benchmarks, i.e., EgoSchema, MLVU, LVBench, MVBench and Video-MME, demonstrate the state-of-the-art performance and outstanding efficiency of our method. Code is available at https://github.com/IVGSZ/Flash-VStream.

This paper presents a comprehensive exploration of the phenomenon of data redundancy in video understanding, with the aim to improve computational efficiency. Our investigation commences with an examination of spatial redundancy, which refers to the observation that the most informative region in each video frame usually corresponds to a small image patch, whose shape, size and location shift smoothly across frames. Motivated by this phenomenon, we formulate the patch localization problem as a dynamic decision task, and introduce a spatially adaptive video recognition approach, termed AdaFocus. In specific, a lightweight encoder is first employed to quickly process the full video sequence, whose features are then utilized by a policy network to identify the most task-relevant regions. Subsequently, the selected patches are inferred by a high-capacity deep network for the final prediction. The complete model can be trained conveniently in an end-to-end manner. During inference, once the informative patch sequence has been generated, the bulk of computation can be executed in parallel, rendering it efficient on modern GPU devices. Furthermore, we demonstrate that AdaFocus can be easily extended by further considering the temporal and sample- wise redundancies, i.e., allocating the majority of computation to the most task-relevant video frames, and minimizing the computation spent on relatively "easier" videos. Our resulting algorithm, Uni-AdaFocus, establishes a comprehensive framework that seamlessly integrates spatial, temporal, and sample- wise dynamic computation, while it preserves the merits of AdaFocus in terms of efficient end-to-end training and hardware friendliness. In addition, Uni-AdaFocus is general and flexible as it is compatible with off-the-shelf backbone models (e.g., TSM and X3D), which can be readily deployed as our feature extractor, yielding a significantly improved computational efficiency. Empirically, extensive experiments based on seven widely-used benchmark datasets (i.e., ActivityNet, FCVID, Mini-Kinetics, Something-Something V1&V2, Jester, and Kinetics-400) and three real-world application scenarios (i.e., fine-grained diving action classification, Alzheimer's and Parkinson's diseases diagnosis with brain magnetic resonance images (MRI), and violence recognition for online videos) substantiate that Uni-AdaFocus is considerably more efficient than the competitive baselines. Code and pre-trained models are available at https://github.com/blackfeather-wang/AdaFocus and https://github.com/LeapLabTHU/AdaFocusV2.

Recent advancements in pre-trained vision-language models like CLIP, have enabled the task of open-vocabulary segmentation. CLIP demonstrates impressive zero-shot capabilities in various downstream tasks that require holistic image understanding. However, due to the image-level contrastive learning and fully global feature interaction, ViT-based CLIP struggles to capture local details, resulting in poor performance in segmentation tasks. Our analysis of ViT-based CLIP reveals that anomaly tokens emerge during the forward process, attracting disproportionate attention from normal patch tokens and thereby diminishing spatial awareness. To address this issue, we propose Self-Calibrated CLIP (SC-CLIP), a training-free method that calibrates CLIP to generate finer representations while preserving its original generalization ability-without introducing new parameters or relying on additional backbones. Specifically, we mitigate the negative impact of anomaly tokens from two complementary perspectives. First, we explicitly identify the anomaly tokens and replace them based on local context. Second, we reduce their influence on normal tokens by enhancing feature discriminability and attention correlation, leveraging the inherent semantic consistency within CLIP's mid-level features. In addition, we introduce a two-pass strategy that effectively integrates multi-level features to enrich local details under the training-free setting. Together, these strategies enhance CLIP's feature representations with improved granularity and semantic coherence. Experimental results demonstrate the effectiveness of SC-CLIP, achieving state-of-the-art results across all datasets and surpassing previous methods by 9.5%. Notably, SC-CLIP boosts the performance of vanilla CLIP ViT-L/14 by 6.8 times. Furthermore, we discuss our method's applicability to other vision-language models and tasks for a comprehensive evaluation. Our source code is available at https://github.com/SuleBai/SC-CLIP.

We study the 3D-aware image attribute editing problem in this paper, which has wide applications in practice. Recent methods solved the problem by training a shared encoder to map images into a 3D generator's latent space or by per-image latent code optimization and then edited images in the latent space. Despite their promising results near the input view, they still suffer from the 3D inconsistency of produced images at large camera poses and imprecise image attribute editing, like affecting unspecified attributes during editing. For more efficient image inversion, we train a shared encoder for all images. To alleviate 3D inconsistency at large camera poses, we propose two novel methods, an alternating training scheme and a multi-view identity loss, to maintain 3D consistency and subject identity. As for imprecise image editing, we attribute the problem to the gap between the latent space of real images and that of generated images. We compare the latent space and inversion manifold of GAN models and demonstrate that editing in the inversion manifold can achieve better results in both quantitative and qualitative evaluations. Extensive experiments show that our method produces more 3D consistent images and achieves more precise image editing than previous work. Source code and pretrained models can be found on our project page: https://mybabyyh.github.io/Preim3D/.