Recent compositional zero-shot learning (CZSL) methods adapt pre-trained vision-language models (VLMs) by constructing trainable prompts only for composed state-object pairs. Relying on learning the joint representation of seen compositions, these methods ignore the explicit modeling of the state and object, thus limiting the exploitation of pre-trained knowledge and generalization to unseen compositions. With a particular focus on the universality of the solution, in this work, we propose a novel paradigm for CZSL models that establishes three identification branches (i.e., Multi-Path) to jointly model the state, object, and composition. The presented Troika is an outstanding implementation that aligns the branch-specific prompt representations with decomposed visual features. To calibrate the bias between semantically similar multi-modal representations, we further devise a Cross-Modal Traction module into Troika that shifts the prompt representation towards the current visual content. We conduct extensive experiments on three popular benchmarks, where our method significantly outperforms existing methods in both closed-world and open-world settings. The code will be available at https://github.com/bighuang624/Troika.

Despite encouraging results from recent developments in transfer learning for adapting pre-trained model to downstream tasks, the performance of model probing is still lagging behind the state-of-the-art parameter efficient tuning methods. Our investigation reveals that existing model probing methods perform well for the easy case when the source domain (where models are pre-trained) and the adapted domain are similar, but fail for the difficult case when the two domains are significantly different. Simply incorporating features extracted from multiple layers and increasing complexity of the probing model can mitigate the gap in the difficult case, but degrades the performance in the easy case. To address this challenge, we propose structured model probing (SMP) that is able to deliver good performance for both cases through structured regularization. The regularization performs feature selection leveraging model structure as a prior, and controls the complexity of the probing model through the weights of selected structures. This enables us to construct a simple adaptation model, with a small number of selected features and a linear prediction model, for the easy case; and to automatically increase the complexity of adaptation model, with a large number of selected features and a non-linear model, for the difficult case. Our extensive empirical studies show that SMP significantly outperforms the state-of-the-art methods for parameter efficient tuning, and at the same time, still maintains the advantage of computational efficiency for probing-based methods.

Standard approaches for video action recognition usually operate on full input videos, which is inefficient due to the widespread spatio-temporal redundancy in videos. The recent progress in masked video modelling, specifically VideoMAE, has shown the ability of vanilla Vision Transformers (ViT) to complement spatio-temporal contexts using limited visual content. Inspired by this, we propose Masked Action Recognition (MAR), which reduces redundant computation by discarding a proportion of patches and operating only on a portion of the videos. MAR includes two essential components:cell running maskingandbridging classifier. Specifically, to enable the ViT to perceive the details beyond the visible patches, cell running masking is used to preserve the spatio-temporal correlations in videos. This ensures that the patches at the same spatial location can be observed in turn for easy reconstructions. Additionally, we notice that, although the partially observed features can reconstruct semantically explicit invisible patches, they fail to achieve accurate classification. To address this issue, we propose a bridging classifier that can help fill the semantic gap between the ViT encoded features used for reconstruction and the specialized features used for classification. Our proposed MAR can reduce the computational cost of ViT by 53%. Extensive experiments have demonstrated that MAR consistently outperforms existing ViT models by a notable margin. Notably, we found that a ViT-Large model fine-tuned by MAR achieves comparable performance to a ViT-Huge model fine-tuned by standard training methods on both Kinetics-400 and Something-Something v2 datasets. Moreover, the computation overhead of our ViT-Large model is only 14.5% of that of the ViT-Huge model. Codes have been made availablehttps://github.com/alibaba-mmai-research/Masked-Action-Recognition.

Many recent studies leverage the pre-trained CLIP for text-video cross-modal retrieval by tuning the backbone with additional heavy modules, which not only brings huge computational burdens with much more parameters, but also leads to the knowledge forgetting from upstream models. In this work, we propose the VoP: Text-Video Cooperative Prompt Tuning for efficient tuning on the text-video retrieval task. The proposed VoP is an end-to-end framework with both video & text prompts introducing, which can be regarded as a powerful baseline with only 0.1% trainable parameters. Further, based on the spatiotemporal characteristics of videos, we develop three novel video prompt mechanisms to improve the performance with different scales of trainable parameters. The basic idea of the VoP enhancement is to model the frame position, frame context, and layer function with specific trainable prompts, respectively. Extensive experiments show that compared to full fine-tuning, the enhanced VoP achieves a 1.4% average R@1 gain across five text-video retrieval benchmarks with 6× less parameter overhead. The code will be available at https://github.com/bighuang624/VoP.

Foundation models are pre-trained on massive data and transferred to downstream tasks via fine-tuning. This work presents Vision Middleware (ViM), a new learning paradigm that targets unified transferring from a single foundation model to a variety of downstream tasks. ViM consists of a zoo of lightweight plug-in modules, each of which is independently learned on a midstream dataset with a shared frozen backbone. Downstream tasks can then benefit from an adequate aggregation of the module zoo thanks to the rich knowledge inherited from midstream tasks. There are three major advantages of such a design. From the efficiency aspect, the upstream backbone can be trained only once and reused for all downstream tasks without tuning. From the scalability aspect, we can easily append additional modules to ViM with no influence on existing modules. From the performance aspect, ViM can include as many midstream tasks as possible, narrowing the task gap between upstream and downstream. Considering these benefits, we believe that ViM, which the community could maintain and develop together, would serve as a powerful tool to assist foundation models.

Video temporal grounding aims to pinpoint a video segment that matches the query description. Despite the recent advance in short-form videos (e.g., in minutes), temporal grounding in long videos (e.g., in hours) is still at its early stage. To address this challenge, a common practice is to employ a sliding window, yet can be inefficient and inflexible due to the limited number of frames within the window. In this work, we propose an end-to-end framework for fast temporal grounding, which is able to model an hours-long video with one-time network execution. Our pipeline is formulated in a coarse-to-fine manner, where we first extract context knowledge from non-overlapped video clips (i.e., anchors), and then supplement the anchors that highly response to the query with detailed content knowledge. Besides the remarkably high pipeline efficiency, another advantage of our approach is the capability of capturing long-range temporal correlation, thanks to modeling the entire video as a whole, and hence facilitates more accurate grounding. Experimental results suggest that, on the long-form video datasets MAD and Ego4d, our method significantly outperforms state-of-the-arts, and achieves 14.6× / 102.8× higher efficiency respectively. Project can be found at https://github.com/afcedf/SOONet.git.

Parameter-efficient tuning has become a trend in transferring large-scale foundation models to downstream applications. Existing methods typically embed some light-weight tuners into the backbone, where both the design and the learning of the tuners are highly dependent on the base model. This work offers a new tuning paradigm, dubbed Res-Tuning, which intentionally unbinds tuners from the backbone. With both theoretical and empirical evidence, we show that popular tuning approaches have their equivalent counterparts under our unbinding formulation, and hence can be integrated into our framework effortlessly. Thanks to the structural disentanglement, we manage to free the design of tuners from the network architecture, facilitating flexible combination of various tuning strategies. We further propose a memory-efficient variant of Res-Tuning, where the bypass i.e., formed by a sequence of tuners) is effectively detached from the main branch, such that the gradients are back-propagated only to the tuners but not to the backbone. Such a detachment also allows one-time backbone forward for multi-task inference. Extensive experiments on both discriminative and generative tasks demonstrate the superiority of our method over existing alternatives from the perspectives of efficacy and efficiency. Project page: https://res-tuning.github.io/.

Near-duplicate video retrieval (NDVR) aims to find the copies or transformations of the query video from a massive video database. It plays an important role in many video related applications, including copyright protection, tracing, filtering and etc. Video representation and similarity search are crucial to any video retrieval system. To derive effective video representation, most video retrieval systems require a large amount of manually annotated data for training, making it costly inefficient. In addition, most retrieval systems are based on frame-level features for video similarity searching, making it expensive both storage wise and search wise. To address the above issues, we propose a video representation learning (VRL) approach to effectively address the above shortcomings. It first effectively learns video representation from unlabeled videos via contrastive learning to avoid the expensive cost of manual annotation. Then, it exploits transformer structure to aggregate frame-level features into clip-level to reduce both storage space and search complexity. It can learn the complementary and discriminative information from the interactions among clip frames, as well as acquire the frame permutation and missing invariant ability to support more flexible retrieval manners. Comprehensive experiments on two challenging near-duplicate video retrieval datasets, namely FIVR-200K and SVD, verify the effectiveness of our proposed VRL approach, which achieves the best performance of video retrieval on accuracy and efficiency.

Existing video copy detection methods generally measure video similarity based on spatial similarities between key frames, neglecting the latent similarity in temporal dimension, so that the video similarity is biased towards spatial information. There are methods modeling unified video similarity in an end-to-end way, but losing detailed partial alignment information, which causes the incapability of copy segments localization. To address the above issues, we propose the Video Similarity and Alignment Learning (VSAL) approach, which jointly models spatial similarity, temporal similarity and partial alignment. To mitigate the spatial similarity bias, we model the temporal similarity as the mask map predicted from frame-level spatial similarity, where each element indicates the probability of frame pair lying right on the partial alignments. To further localize partial copies, the step map is learned from the spatial similarity where the elements indicate extending directions of the current partial alignments on the spatial-temporal similarity map. Obtained from the mask map, the start points extend out into partial optimal alignments following instructions of the step map. With the similarity and alignment learning strategy, VSAL achieves the state-of-the-art F1-score on VCDB core dataset. Furthermore, we construct a new benchmark of partial video copy detection and localization by adding new segment-level annotations for FIVR-200k dataset, where VSAL also achieves the best performance, verifying its effectiveness in more challenging situations. Our project is publicly available at https://pvcd-vsal.github.io/vsal/.

Video-text retrieval is an important yet challenging task in vision-language understanding, which aims to learn a joint embedding space where related video and text instances are close to each other. Most current works simply measure the video-text similarity based on video-level and text-level embeddings. However, the neglect of more fine-grained or local information causes the problem of insufficient representation. Some works exploit the local details by disentangling sentences, but overlook the corresponding videos, causing the asymmetry of video-text representation. To address the above limitations, we propose a Hierarchical Alignment Network (HANet) to align different level representations for video-text matching. Specifically, we first decompose video and text into three semantic levels, namely event (video and text), action (motion and verb), and entity (appearance and noun). Based on these, we naturally construct hierarchical representations in the individual-local-global manner, where the individual level focuses on the alignment between frame and word, local level focuses on the alignment between video clip and textual context, and global level focuses on the alignment between the whole video and text. Different level alignments capture fine-to-coarse correlations between video and text, as well as take the advantage of the complementary information among three semantic levels. Besides, our HANet is also richly interpretable by explicitly learning key semantic concepts. Extensive experiments on two public datasets, namely MSR-VTT and VATEX, show the proposed HANet outperforms other state-of-the-art methods, which demonstrates the effectiveness of hierarchical representation and alignment. Our code is publicly available at https://github.com/Roc-Ng/HANet.