Accelerating Pre-training of Multimodal LLMs via Chain-of-Sight

This paper introduces Chain-of-Sight, a vision-language bridge module that accelerates the pre-training of Multimodal Large Language Models (MLLMs). Our approach employs a sequence of visual resamplers that capture visual details at various spacial scales. This architecture not only leverages global and local visual contexts effectively, but also facilitates the flexible extension of visual tokens through a compound token scaling strategy, allowing up to a 16x increase in the token count post pre-training. Consequently, Chain-of-Sight requires significantly fewer visual tokens in the pre-training phase compared to the fine-tuning phase. This intentional reduction of visual tokens during pre-training notably accelerates the pre-training process, cutting down the wall-clock training time by ~73%. Empirical results on a series of vision-language benchmarks reveal that the pre-train acceleration through Chain-of-Sight is achieved without sacrificing performance, matching or surpassing the standard pipeline of utilizing all visual tokens throughout the entire training process. Further scaling up the number of visual tokens for pre-training leads to stronger performances, competitive to existing approaches in a series of benchmarks.

A Recipe for Scaling up Text-to-Video Generation with Text-free Videos

Diffusion-based text-to-video generation has witnessed impressive progress in the past year yet still falls behind text-to-image generation. One of the key reasons is the limited scale of publicly available data (e.g., 10M video-text pairs in WebVid10M vs. 5B image-text pairs in LAION), considering the high cost of video captioning. Instead, it could be far easier to collect unlabeled clips from video platforms like YouTube. Motivated by this, we come up with a novel text-to-video generation framework, termed TF-T2V, which can directly learn with text-free videos. The rationale behind is to separate the process of text decoding from that of temporal modeling. To this end, we employ a content branch and a motion branch, which are jointly optimized with weights shared. Following such a pipeline, we study the effect of doubling the scale of training set (i.e., video-only WebVid10M) with some randomly collected text-free videos and are encouraged to observe the performance improvement (FID from 9.67 to 8.19 and FVD from 484 to 441), demonstrating the scalability of our approach. We also find that our model could enjoy sustainable performance gain (FID from 8.19 to 7.64 and FVD from 441 to 366) after reintroducing some text labels for training. Finally, we validate the effectiveness and generalizability of our ideology on both native text-to-video generation and compositional video synthesis paradigms. Code and models will be publicly available at https://tf-t2v.github.io/.

Hierarchical Spatio-temporal Decoupling for Text-to-Video Generation

Despite diffusion models having shown powerful abilities to generate photorealistic images, generating videos that are realistic and diverse still remains in its infancy. One of the key reasons is that current methods intertwine spatial content and temporal dynamics together, leading to a notably increased complexity of text-to-video generation (T2V). In this work, we propose HiGen, a diffusion model-based method that improves performance by decoupling the spatial and temporal factors of videos from two perspectives, i.e., structure level and content level. At the structure level, we decompose the T2V task into two steps, including spatial reasoning and temporal reasoning, using a unified denoiser. Specifically, we generate spatially coherent priors using text during spatial reasoning and then generate temporally coherent motions from these priors during temporal reasoning. At the content level, we extract two subtle cues from the content of the input video that can express motion and appearance changes, respectively. These two cues then guide the model's training for generating videos, enabling flexible content variations and enhancing temporal stability. Through the decoupled paradigm, HiGen can effectively reduce the complexity of this task and generate realistic videos with semantics accuracy and motion stability. Extensive experiments demonstrate the superior performance of HiGen over the state-of-the-art T2V methods.

DreamVideo: Composing Your Dream Videos with Customized Subject and Motion

Customized generation using diffusion models has made impressive progress in image generation, but remains unsatisfactory in the challenging video generation task, as it requires the controllability of both subjects and motions. To that end, we present DreamVideo, a novel approach to generating personalized videos from a few static images of the desired subject and a few videos of target motion. DreamVideo decouples this task into two stages, subject learning and motion learning, by leveraging a pre-trained video diffusion model. The subject learning aims to accurately capture the fine appearance of the subject from provided images, which is achieved by combining textual inversion and fine-tuning of our carefully designed identity adapter. In motion learning, we architect a motion adapter and fine-tune it on the given videos to effectively model the target motion pattern. Combining these two lightweight and efficient adapters allows for flexible customization of any subject with any motion. Extensive experimental results demonstrate the superior performance of our DreamVideo over the state-of-the-art methods for customized video generation. Our project page is at https://dreamvideo-t2v.github.io.

Disentangling Spatial and Temporal Learning for Efficient Image-to-Video Transfer Learning

Recently, large-scale pre-trained language-image models like CLIP have shown extraordinary capabilities for understanding spatial contents, but naively transferring such models to video recognition still suffers from unsatisfactory temporal modeling capabilities. Existing methods insert tunable structures into or in parallel with the pre-trained model, which either requires back-propagation through the whole pre-trained model and is thus resource-demanding, or is limited by the temporal reasoning capability of the pre-trained structure. In this work, we present DiST, which disentangles the learning of spatial and temporal aspects of videos. Specifically, DiST uses a dual-encoder structure, where a pre-trained foundation model acts as the spatial encoder, and a lightweight network is introduced as the temporal encoder. An integration branch is inserted between the encoders to fuse spatio-temporal information. The disentangled spatial and temporal learning in DiST is highly efficient because it avoids the back-propagation of massive pre-trained parameters. Meanwhile, we empirically show that disentangled learning with an extra network for integration benefits both spatial and temporal understanding. Extensive experiments on five benchmarks show that DiST delivers better performance than existing state-of-the-art methods by convincing gaps. When pre-training on the large-scale Kinetics-710, we achieve 89.7% on Kinetics-400 with a frozen ViT-L model, which verifies the scalability of DiST. Codes and models can be found in https://github.com/alibaba-mmai-research/DiST.

HR-Pro: Point-supervised Temporal Action Localization via Hierarchical Reliability Propagation

Point-supervised Temporal Action Localization (PSTAL) is an emerging research direction for label-efficient learning. However, current methods mainly focus on optimizing the network either at the snippet-level or the instance-level, neglecting the inherent reliability of point annotations at both levels. In this paper, we propose a Hierarchical Reliability Propagation (HR-Pro) framework, which consists of two reliability-aware stages: Snippet-level Discrimination Learning and Instance-level Completeness Learning, both stages explore the efficient propagation of high-confidence cues in point annotations. For snippet-level learning, we introduce an online-updated memory to store reliable snippet prototypes for each class. We then employ a Reliability-aware Attention Block to capture both intra-video and inter-video dependencies of snippets, resulting in more discriminative and robust snippet representation. For instance-level learning, we propose a point-based proposal generation approach as a means of connecting snippets and instances, which produces high-confidence proposals for further optimization at the instance level. Through multi-level reliability-aware learning, we obtain more reliable confidence scores and more accurate temporal boundaries of predicted proposals. Our HR-Pro achieves state-of-the-art performance on multiple challenging benchmarks, including an impressive average mAP of 60.3% on THUMOS14. Notably, our HR-Pro largely surpasses all previous point-supervised methods, and even outperforms several competitive fully supervised methods. Code will be available at https://github.com/pipixin321/HR-Pro.

Temporally-Adaptive Models for Efficient Video Understanding

Spatial convolutions are extensively used in numerous deep video models. It fundamentally assumes spatio-temporal invariance, i.e., using shared weights for every location in different frames. This work presents Temporally-Adaptive Convolutions (TAdaConv) for video understanding, which shows that adaptive weight calibration along the temporal dimension is an efficient way to facilitate modeling complex temporal dynamics in videos. Specifically, TAdaConv empowers spatial convolutions with temporal modeling abilities by calibrating the convolution weights for each frame according to its local and global temporal context. Compared to existing operations for temporal modeling, TAdaConv is more efficient as it operates over the convolution kernels instead of the features, whose dimension is an order of magnitude smaller than the spatial resolutions. Further, kernel calibration brings an increased model capacity. Based on this readily plug-in operation TAdaConv as well as its extension, i.e., TAdaConvV2, we construct TAdaBlocks to empower ConvNeXt and Vision Transformer to have strong temporal modeling capabilities. Empirical results show TAdaConvNeXtV2 and TAdaFormer perform competitively against state-of-the-art convolutional and Transformer-based models in various video understanding benchmarks. Our codes and models are released at: https://github.com/alibaba-mmai-research/TAdaConv.

MoLo: Motion-augmented Long-short Contrastive Learning for Few-shot Action Recognition

Current state-of-the-art approaches for few-shot action recognition achieve promising performance by conducting frame-level matching on learned visual features. However, they generally suffer from two limitations: i) the matching procedure between local frames tends to be inaccurate due to the lack of guidance to force long-range temporal perception; ii) explicit motion learning is usually ignored, leading to partial information loss. To address these issues, we develop a Motion-augmented Long-short Contrastive Learning (MoLo) method that contains two crucial components, including a long-short contrastive objective and a motion autodecoder. Specifically, the long-short contrastive objective is to endow local frame features with long-form temporal awareness by maximizing their agreement with the global token of videos belonging to the same class. The motion autodecoder is a lightweight architecture to reconstruct pixel motions from the differential features, which explicitly embeds the network with motion dynamics. By this means, MoLo can simultaneously learn long-range temporal context and motion cues for comprehensive few-shot matching. To demonstrate the effectiveness, we evaluate MoLo on five standard benchmarks, and the results show that MoLo favorably outperforms recent advanced methods. The source code is available at https://github.com/alibaba-mmai-research/MoLo.

Enlarging Instance-specific and Class-specific Information for Open-set Action Recognition

Open-set action recognition is to reject unknown human action cases which are out of the distribution of the training set. Existing methods mainly focus on learning better uncertainty scores but dismiss the importance of feature representations. We find that features with richer semantic diversity can significantly improve the open-set performance under the same uncertainty scores. In this paper, we begin with analyzing the feature representation behavior in the open-set action recognition (OSAR) problem based on the information bottleneck (IB) theory, and propose to enlarge the instance-specific (IS) and class-specific (CS) information contained in the feature for better performance. To this end, a novel Prototypical Similarity Learning (PSL) framework is proposed to keep the instance variance within the same class to retain more IS information. Besides, we notice that unknown samples sharing similar appearances to known samples are easily misclassified as known classes. To alleviate this issue, video shuffling is further introduced in our PSL to learn distinct temporal information between original and shuffled samples, which we find enlarges the CS information. Extensive experiments demonstrate that the proposed PSL can significantly boost both the open-set and closed-set performance and achieves state-of-the-art results on multiple benchmarks. Code is available at https://github.com/Jun-CEN/PSL.

HyRSM++: Hybrid Relation Guided Temporal Set Matching for Few-shot Action Recognition

Recent attempts mainly focus on learning deep representations for each video individually under the episodic meta-learning regime and then performing temporal alignment to match query and support videos. However, they still suffer from two drawbacks: (i) learning individual features without considering the entire task may result in limited representation capability, and (ii) existing alignment strategies are sensitive to noises and misaligned instances. To handle the two limitations, we propose a novel Hybrid Relation guided temporal Set Matching (HyRSM++) approach for few-shot action recognition. The core idea of HyRSM++ is to integrate all videos within the task to learn discriminative representations and involve a robust matching technique. To be specific, HyRSM++ consists of two key components, a hybrid relation module and a temporal set matching metric. Given the basic representations from the feature extractor, the hybrid relation module is introduced to fully exploit associated relations within and cross videos in an episodic task and thus can learn task-specific embeddings. Subsequently, in the temporal set matching metric, we carry out the distance measure between query and support videos from a set matching perspective and design a Bi-MHM to improve the resilience to misaligned instances. In addition, we explicitly exploit the temporal coherence in videos to regularize the matching process. Furthermore, we extend the proposed HyRSM++ to deal with the more challenging semi-supervised few-shot action recognition and unsupervised few-shot action recognition tasks. Experimental results on multiple benchmarks demonstrate that our method achieves state-of-the-art performance under various few-shot settings. The source code is available at https://github.com/alibaba-mmai-research/HyRSMPlusPlus.