Video representation learning has been successful in video-text pre-training for zero-shot transfer, where each sentence is trained to be close to the paired video clips in a common feature space. For long videos, given a paragraph of description where the sentences describe different segments of the video, by matching all sentence-clip pairs, the paragraph and the full video are aligned implicitly. However, such unit-level similarity measure may ignore the global temporal context over a long time span, which inevitably limits the generalization ability. In this paper, we propose a contrastive learning framework TempCLR to compare the full video and the paragraph explicitly. As the video/paragraph is formulated as a sequence of clips/sentences, under the constraint of their temporal order, we use dynamic time warping to compute the minimum cumulative cost over sentence-clip pairs as the sequence-level distance. To explore the temporal dynamics, we break the consistency of temporal order by shuffling the video clips or sentences according to the temporal granularity. In this way, we obtain the representations for clips/sentences, which perceive the temporal information and thus facilitate the sequence alignment. In addition to pre-training on the video and paragraph, our approach can also generalize on the matching between different video instances. We evaluate our approach on video retrieval, action step localization, and few-shot action recognition, and achieve consistent performance gain over all three tasks. Detailed ablation studies are provided to justify the approach design.
From a visual scene containing multiple people, human is able to distinguish each individual given the context descriptions about what happened before, their mental/physical states or intentions, etc. Above ability heavily relies on human-centric commonsense knowledge and reasoning. For example, if asked to identify the "person who needs healing" in an image, we need to first know that they usually have injuries or suffering expressions, then find the corresponding visual clues before finally grounding the person. We present a new commonsense task, Human-centric Commonsense Grounding, that tests the models' ability to ground individuals given the context descriptions about what happened before, and their mental/physical states or intentions. We further create a benchmark, HumanCog, a dataset with 130k grounded commonsensical descriptions annotated on 67k images, covering diverse types of commonsense and visual scenes. We set up a context-object-aware method as a strong baseline that outperforms previous pre-trained and non-pretrained models. Further analysis demonstrates that rich visual commonsense and powerful integration of multi-modal commonsense are essential, which sheds light on future works. Data and code will be available https://github.com/Hxyou/HumanCog.
Visual commonsense understanding requires Vision Language (VL) models to not only understand image and text but also cross-reference in-between to fully integrate and achieve comprehension of the visual scene described. Recently, various approaches have been developed and have achieved high performance on visual commonsense benchmarks. However, it is unclear whether the models really understand the visual scene and underlying commonsense knowledge due to limited evaluation data resources. To provide an in-depth analysis, we present a Multimodal Evaluation (ME) pipeline to automatically generate question-answer pairs to test models' understanding of the visual scene, text, and related knowledge. We then take a step further to show that training with the ME data boosts the model's performance in standard VCR evaluation. Lastly, our in-depth analysis and comparison reveal interesting findings: (1) semantically low-level information can assist the learning of high-level information but not the opposite; (2) visual information is generally under utilization compared with text.
Video event extraction aims to detect salient events from a video and identify the arguments for each event as well as their semantic roles. Existing methods focus on capturing the overall visual scene of each frame, ignoring fine-grained argument-level information. Inspired by the definition of events as changes of states, we propose a novel framework to detect video events by tracking the changes in the visual states of all involved arguments, which are expected to provide the most informative evidence for the extraction of video events. In order to capture the visual state changes of arguments, we decompose them into changes in pixels within objects, displacements of objects, and interactions among multiple arguments. We further propose Object State Embedding, Object Motion-aware Embedding and Argument Interaction Embedding to encode and track these changes respectively. Experiments on various video event extraction tasks demonstrate significant improvements compared to state-of-the-art models. In particular, on verb classification, we achieve 3.49% absolute gains (19.53% relative gains) in F1@5 on Video Situation Recognition.
In Video Question Answering (VideoQA), answering general questions about a video requires its visual information. Yet, video often contains redundant information irrelevant to the VideoQA task. For example, if the task is only to answer questions similar to "Is someone laughing in the video?", then all other information can be discarded. This paper investigates how many bits are really needed from the video in order to do VideoQA by introducing a novel Few-Bit VideoQA problem, where the goal is to accomplish VideoQA with few bits of video information (e.g., 10 bits). We propose a simple yet effective task-specific feature compression approach to solve this problem. Specifically, we insert a lightweight Feature Compression Module (FeatComp) into a VideoQA model which learns to extract task-specific tiny features as little as 10 bits, which are optimal for answering certain types of questions. We demonstrate more than 100,000-fold storage efficiency over MPEG4-encoded videos and 1,000-fold over regular floating point features, with just 2.0-6.6% absolute loss in accuracy, which is a surprising and novel finding. Finally, we analyze what the learned tiny features capture and demonstrate that they have eliminated most of the non-task-specific information, and introduce a Bit Activation Map to visualize what information is being stored. This decreases the privacy risk of data by providing k-anonymity and robustness to feature-inversion techniques, which can influence the machine learning community, allowing us to store data with privacy guarantees while still performing the task effectively.
Large-scale multi-modal contrastive pre-training has demonstrated great utility to learn transferable features for a range of downstream tasks by mapping multiple modalities into a shared embedding space. Typically, this has employed separate encoders for each modality. However, recent work suggests that transformers can support learning across multiple modalities and allow knowledge sharing. Inspired by this, we investigate a variety of Modality-Shared Contrastive Language-Image Pre-training (MS-CLIP) frameworks. More specifically, we question how many parameters of a transformer model can be shared across modalities during contrastive pre-training, and rigorously examine architectural design choices that position the proportion of parameters shared along a spectrum. In studied conditions, we observe that a mostly unified encoder for vision and language signals outperforms all other variations that separate more parameters. Additionally, we find that light-weight modality-specific parallel modules further improve performance. Experimental results show that the proposed MS-CLIP approach outperforms vanilla CLIP by up to 13\% relative in zero-shot ImageNet classification (pre-trained on YFCC-100M), while simultaneously supporting a reduction of parameters. In addition, our approach outperforms vanilla CLIP by 1.6 points in linear probing on a collection of 24 downstream vision tasks. Furthermore, we discover that sharing parameters leads to semantic concepts from different modalities being encoded more closely in the embedding space, facilitating the transferring of common semantic structure (e.g., attention patterns) from language to vision. Code is available at \href{https://github.com/Hxyou/MSCLIP}{URL}.
Understanding how events described or shown in multimedia content relate to one another is a critical component to developing robust artificially intelligent systems which can reason about real-world media. While much research has been devoted to event understanding in the text, image, and video domains, none have explored the complex relations that events experience across domains. For example, a news article may describe a `protest' event while a video shows an `arrest' event. Recognizing that the visual `arrest' event is a subevent of the broader `protest' event is a challenging, yet important problem that prior work has not explored. In this paper, we propose the novel task of MultiModal Event Event Relations to recognize such cross-modal event relations. We contribute a large-scale dataset consisting of 100k video-news article pairs, as well as a benchmark of densely annotated data. We also propose a weakly supervised multimodal method which integrates commonsense knowledge from an external knowledge base (KB) to predict rich multimodal event hierarchies. Experiments show that our model outperforms a number of competitive baselines on our proposed benchmark. We also perform a detailed analysis of our model's performance and suggest directions for future research.
Multi-channel video-language retrieval require models to understand information from different modalities (e.g. video+question, video+speech) and real-world knowledge to correctly link a video with a textual response or query. Fortunately, multimodal contrastive models have been shown to be highly effective at aligning entities in images/videos and text, e.g., CLIP; text contrastive models have been extensively studied recently for their strong ability of producing discriminative sentence embeddings, e.g., SimCSE. Their abilities are exactly needed by multi-channel video-language retrieval. However, it is not clear how to quickly adapt these two lines of models to multi-channel video-language retrieval-style tasks. In this paper, we identify a principled model design space with two axes: how to represent videos and how to fuse video and text information. Based on categorization of recent methods, we investigate the options of representing videos using continuous feature vectors or discrete text tokens; for the fusion method, we explore a multimodal transformer or a pretrained contrastive text model. We extensively evaluate the four combinations on five video-language datasets. We surprisingly find that discrete text tokens coupled with a pretrained contrastive text model yields the best performance. This combination can even outperform state-of-the-art on the iVQA dataset without the additional training on millions of video-language data. Further analysis shows that this is because representing videos as text tokens captures the key visual information with text tokens that are naturally aligned with text models and the text models obtained rich knowledge during contrastive pretraining process. All the empirical analysis we obtain for the four variants establishes a solid foundation for future research on leveraging the rich knowledge of pretrained contrastive models.
The goal of this work is to build flexible video-language models that can generalize to various video-to-text tasks from few examples, such as domain-specific captioning, question answering, and future event prediction. Existing few-shot video-language learners focus exclusively on the encoder, resulting in the absence of a video-to-text decoder to handle generative tasks. Video captioners have been pretrained on large-scale video-language datasets, but they rely heavily on finetuning and lack the ability to generate text for unseen tasks in a few-shot setting. We propose VidIL, a few-shot Video-language Learner via Image and Language models, which demonstrates strong performance on few-shot video-to-text tasks without the necessity of pretraining or finetuning on any video datasets. We use the image-language models to translate the video content into frame captions, object, attribute, and event phrases, and compose them into a temporal structure template. We then instruct a language model, with a prompt containing a few in-context examples, to generate a target output from the composed content. The flexibility of prompting allows the model to capture any form of text input, such as automatic speech recognition (ASR) transcripts. Our experiments demonstrate the power of language models in understanding videos on a wide variety of video-language tasks, including video captioning, video question answering, video caption retrieval, and video future event prediction. Especially, on video future event prediction, our few-shot model significantly outperforms state-of-the-art supervised models trained on large-scale video datasets. Code and resources are publicly available for research purposes at https://github.com/MikeWangWZHL/VidIL .