Test-Time Adaptation (TTA) has recently emerged as a promising research direction, enabling vision-language models (VLMs) to adapt to unlabeled test data in zero-shot settings. Among TTA approaches, test-time prompt tuning has shown great potential for enhancing the practical applicability of VLMs. However, existing methods typically either focus on adapting a single modality or apply uniform optimization to both modalities, without explicitly defining modality-specific optimization objectives. Such a one-size-fits-all strategy often results in suboptimal performance under test-time conditions. To address this limitation, we propose Dual-modality Heterogeneous Prompt Tuning (DHPT), a novel framework designed to simultaneously capture fine-grained textual semantics and alleviate domain shift noise in the visual modality. Specifically, we leverage a large language model to provide textual cognition guidance for the text encoder, while on the vision side, we develop a lightweight calibration module that adaptively mitigates domain shift noise across different scales. Furthermore, we introduce a cluster-tight optimization objective that enhances the stability and generalizability of prompt tuning under distribution shifts. Extensive experiments conducted on 11 benchmark datasets demonstrate that DHPT consistently and significantly outperforms existing TTA methods for VLMs.

Transformers are increasingly becoming the dominant model in the field of computer vision, thereby catalyzing research efforts aimed at unraveling the interpretability of transformers. Existing explanation techniques, whether attention-based or gradient-based, furnish a dependable approach to quantifying the impact of input features on model predictions from the perspective of dissecting self-attention mechanisms. However, current research overlooks the block-to-block constraints, which result in misdirection in attribution. In this work, we propose a block-wise constraints-free interpretation method, Independent Block Level Attribution (IBA), which maintains the relative independence of each block in the model. The IBA reconfigures the model into mutually unaffected class-semantic blocks via class-semantic relevance, each of which performs the attribution computation independently, thus minimizing the influence of inter-block constraints on the model interpretation performance. Extensive perturbation and segmentation experiments unequivocally demonstrate the superiority of our method, showcasing its significant outperformance compared to current interpretation methods. Additionally, we also apply IBA to the text transformer to demonstrate the generalization of our method. Our code is available athttps://github.com/qinanin/TMM-IBA.git.

Knowledge distillation (KD), a technique widely employed in computer vision, has emerged as a de facto standard for improving the performance of small neural networks. However, prevailing KD-based approaches in video tasks primarily focus on designing loss functions and fusing cross-modal information. This overlooks the spatial-temporal feature semantics, resulting in limited advancements in model compression. Addressing this gap, our paper introduces an innovative knowledge distillation framework, with the generative model for training a lightweight student model. In particular, the framework is organized into two steps: the initial phase is Feature Representation, wherein a generative model-based attention module is trained to represent feature semantics; Subsequently, the Generative-based Feature Distillation phase encompasses both Generative Distillation and Attention Distillation, with the objective of transferring attention-based feature semantics with the generative model. The efficacy of our approach is demonstrated through comprehensive experiments on diverse popular datasets, proving considerable enhancements in video action recognition task. Moreover, the effectiveness of our proposed framework is validated in the context of more intricate video action detection task. Our code is available at https://github.com/aaai-24/Generative-based-KD.

Weakly-supervised action localization aims to recognize and localize action instancese in untrimmed videos with only video-level labels. Most existing models rely on multiple instance learning(MIL), where the predictions of unlabeled instances are supervised by classifying labeled bags. The MIL-based methods are relatively well studied with cogent performance achieved on classification but not on localization. Generally, they locate temporal regions by the video-level classification but overlook the temporal variations of feature semantics. To address this problem, we propose a novel attention-based hierarchically-structured latent model to learn the temporal variations of feature semantics. Specifically, our model entails two components, the first is an unsupervised change-points detection module that detects change-points by learning the latent representations of video features in a temporal hierarchy based on their rates of change, and the second is an attention-based classification model that selects the change-points of the foreground as the boundaries. To evaluate the effectiveness of our model, we conduct extensive experiments on two benchmark datasets, THUMOS-14 and ActivityNet-v1.3. The experiments show that our method outperforms current state-of-the-art methods, and even achieves comparable performance with fully-supervised methods.

Video object detection is a fundamental technology of intelligent video analytics for Internet of Things (IoT) applications. However, even with extraordinary detection accuracy, predominating solutions based on deep convolutional neural networks (DCNNs) cannot achieve real-time online object detection on video streams with a low end-to-end (E2E) response latency and therefore cannot be applied to proliferating latency-sensitive IoT applications like autonomous driving requiring large-scale intelligent video analytics. To address this issue, we present EC2Detect, an edge-cloud collaborative real-time online video object detection method. Specifically, we propose a tracking-assisted object detection architecture based on edge-cloud collaboration with keyframe selection, where the accurate but heavy object detection is conducted by the Cloud on sparse keyframes adaptively selected according to their semantic variation, and the lightweight object tracking is used to localize and identify objects in other frames at edge devices. Extensive experiments of our real-world prototype demonstrate that, EC2Detect significantly outperforms state-of-the-art methods in terms of processing speed (up to$4.77\times $faster), E2E latency (up to$8.12 \times $lower), and edge-cloud bandwidth occupation ($17 \times $lower) with an acceptable mAP, which can effectively support large-scale intelligent video analytics in practice. Source code of EC2Detect is available athttps://github.com/ECCDetect/ECCDetect.