Automatic video captioning aims for a holistic visual scene understanding. It requires a mechanism for capturing temporal context in video frames and the ability to comprehend the actions and associations of objects in a given timeframe. Such a system should additionally learn to abstract video sequences into sensible representations as well as to generate natural written language. While the majority of captioning models focus solely on the visual inputs, little attention has been paid to the audiovisual modality. To tackle this issue, we propose a novel two-fold approach. First, we implement a reward-guided KL Divergence to train a video captioning model which is resilient towards token permutations. Second, we utilise a Bi-Modal Hierarchical Reinforcement Learning (BMHRL) Transformer architecture to capture long-term temporal dependencies of the input data as a foundation for our hierarchical captioning module. Using our BMHRL, we show the suitability of the HRL agent in the generation of content-complete and grammatically sound sentences by achieving $4.91$, $2.23$, and $10.80$ in BLEU3, BLEU4, and METEOR scores, respectively on the ActivityNet Captions dataset. Finally, we make our BMHRL framework and trained models publicly available for users and developers at https://github.com/d-rothen/bmhrl.
Video understanding is a growing field and a subject of intense research, which includes many interesting tasks to understanding both spatial and temporal information, e.g., action detection, action recognition, video captioning, video retrieval. One of the most challenging problems in video understanding is dealing with feature extraction, i.e. extract contextual visual representation from given untrimmed video due to the long and complicated temporal structure of unconstrained videos. Different from existing approaches, which apply a pre-trained backbone network as a black-box to extract visual representation, our approach aims to extract the most contextual information with an explainable mechanism. As we observed, humans typically perceive a video through the interactions between three main factors, i.e., the actors, the relevant objects, and the surrounding environment. Therefore, it is very crucial to design a contextual explainable video representation extraction that can capture each of such factors and model the relationships between them. In this paper, we discuss approaches, that incorporate the human perception process into modeling actors, objects, and the environment. We choose video paragraph captioning and temporal action detection to illustrate the effectiveness of human perception based-contextual representation in video understanding. Source code is publicly available at https://github.com/UARK-AICV/Video_Representation.
We explore an efficient approach to establish a foundational video-text model. We present VideoCoCa that maximally reuses a pretrained image-text contrastive captioner (CoCa) model and adapt it to video-text tasks with minimal extra training. While previous works adapt image-text models with various cross-frame fusion modules, we find that the generative attentional pooling and contrastive attentional pooling layers in CoCa are instantly adaptable to flattened frame embeddings, yielding state-of-the-art results on zero-shot video classification and zero-shot text-to-video retrieval. Furthermore, we explore lightweight finetuning on top of VideoCoCa, and achieve strong results on video question-answering and video captioning.
Video paragraph captioning aims to generate a multi-sentence description of an untrimmed video with several temporal event locations in coherent storytelling. Following the human perception process, where the scene is effectively understood by decomposing it into visual (e.g. human, animal) and non-visual components (e.g. action, relations) under the mutual influence of vision and language, we first propose a visual-linguistic (VL) feature. In the proposed VL feature, the scene is modeled by three modalities including (i) a global visual environment; (ii) local visual main agents; (iii) linguistic scene elements. We then introduce an autoregressive Transformer-in-Transformer (TinT) to simultaneously capture the semantic coherence of intra- and inter-event contents within a video. Finally, we present a new VL contrastive loss function to guarantee learnt embedding features are matched with the captions semantics. Comprehensive experiments and extensive ablation studies on ActivityNet Captions and YouCookII datasets show that the proposed Visual-Linguistic Transformer-in-Transform (VLTinT) outperforms prior state-of-the-art methods on accuracy and diversity. Source code is made publicly available at: https://github.com/UARK-AICV/VLTinT.
Video captioning aims to generate natural language sentences that describe the given video accurately. Existing methods obtain favorable generation by exploring richer visual representations in encode phase or improving the decoding ability. However, the long-tailed problem hinders these attempts at low-frequency tokens, which rarely occur but carry critical semantics, playing a vital role in the detailed generation. In this paper, we introduce a novel Refined Semantic enhancement method towards Frequency Diffusion (RSFD), a captioning model that constantly perceives the linguistic representation of the infrequent tokens. Concretely, a Frequency-Aware Diffusion (FAD) module is proposed to comprehend the semantics of low-frequency tokens to break through generation limitations. In this way, the caption is refined by promoting the absorption of tokens with insufficient occurrence. Based on FAD, we design a Divergent Semantic Supervisor (DSS) module to compensate for the information loss of high-frequency tokens brought by the diffusion process, where the semantics of low-frequency tokens is further emphasized to alleviate the long-tailed problem. Extensive experiments indicate that RSFD outperforms the state-of-the-art methods on two benchmark datasets, i.e., MSR-VTT and MSVD, demonstrate that the enhancement of low-frequency tokens semantics can obtain a competitive generation effect. Code is available at https://github.com/lzp870/RSFD.
Training supervised video captioning model requires coupled video-caption pairs. However, for many targeted languages, sufficient paired data are not available. To this end, we introduce the unpaired video captioning task aiming to train models without coupled video-caption pairs in target language. To solve the task, a natural choice is to employ a two-step pipeline system: first utilizing video-to-pivot captioning model to generate captions in pivot language and then utilizing pivot-to-target translation model to translate the pivot captions to the target language. However, in such a pipeline system, 1) visual information cannot reach the translation model, generating visual irrelevant target captions; 2) the errors in the generated pivot captions will be propagated to the translation model, resulting in disfluent target captions. To address these problems, we propose the Unpaired Video Captioning with Visual Injection system (UVC-VI). UVC-VI first introduces the Visual Injection Module (VIM), which aligns source visual and target language domains to inject the source visual information into the target language domain. Meanwhile, VIM directly connects the encoder of the video-to-pivot model and the decoder of the pivot-to-target model, allowing end-to-end inference by completely skipping the generation of pivot captions. To enhance the cross-modality injection of the VIM, UVC-VI further introduces a pluggable video encoder, i.e., Multimodal Collaborative Encoder (MCE). The experiments show that UVC-VI outperforms pipeline systems and exceeds several supervised systems. Furthermore, equipping existing supervised systems with our MCE can achieve 4% and 7% relative margins on the CIDEr scores to current state-of-the-art models on the benchmark MSVD and MSR-VTT datasets, respectively.
Most video-and-language representation learning approaches employ contrastive learning, e.g., CLIP, to project the video and text features into a common latent space according to the semantic similarities of text-video pairs. However, such learned shared latent spaces are not often optimal, and the modality gap between visual and textual representation can not be fully eliminated. In this paper, we propose Expectation-Maximization Contrastive Learning (EMCL) to learn compact video-and-language representations. Specifically, we use the Expectation-Maximization algorithm to find a compact set of bases for the latent space, where the features could be concisely represented as the linear combinations of these bases. Such feature decomposition of video-and-language representations reduces the rank of the latent space, resulting in increased representing power for the semantics. Extensive experiments on three benchmark text-video retrieval datasets prove that our EMCL can learn more discriminative video-and-language representations than previous methods, and significantly outperform previous state-of-the-art methods across all metrics. More encouragingly, the proposed method can be applied to boost the performance of existing approaches either as a jointly training layer or an out-of-the-box inference module with no extra training, making it easy to be incorporated into any existing methods.
Generating consecutive descriptions for videos, i.e., Video Captioning, requires taking full advantage of visual representation along with the generation process. Existing video captioning methods focus on making an exploration of spatial-temporal representations and their relationships to produce inferences. However, such methods only exploit the superficial association contained in the video itself without considering the intrinsic visual commonsense knowledge that existed in a video dataset, which may hinder their capabilities of knowledge cognitive to reason accurate descriptions. To address this problem, we propose a simple yet effective method, called Visual Commonsense-aware Representation Network (VCRN), for video captioning. Specifically, we construct a Video Dictionary, a plug-and-play component, obtained by clustering all video features from the total dataset into multiple clustered centers without additional annotation. Each center implicitly represents a visual commonsense concept in the video domain, which is utilized in our proposed Visual Concept Selection (VCS) to obtain a video-related concept feature. Next, a Conceptual Integration Generation (CIG) is proposed to enhance the caption generation. Extensive experiments on three publicly video captioning benchmarks: MSVD, MSR-VTT, and VATEX, demonstrate that our method reaches state-of-the-art performance, indicating the effectiveness of our method. In addition, our approach is integrated into the existing method of video question answering and improves this performance, further showing the generalization of our method. Source code has been released at https://github.com/zchoi/VCRN.
Annotation of multimedia data by humans is time-consuming and costly, while reliable automatic generation of semantic metadata is a major challenge. We propose a framework to extract semantic metadata from automatically generated video captions. As metadata, we consider entities, the entities' properties, relations between entities, and the video category. We employ two state-of-the-art dense video captioning models with masked transformer (MT) and parallel decoding (PVDC) to generate captions for videos of the ActivityNet Captions dataset. Our experiments show that it is possible to extract entities, their properties, relations between entities, and the video category from the generated captions. We observe that the quality of the extracted information is mainly influenced by the quality of the event localization in the video as well as the performance of the event caption generation.
Recent visuolinguistic pre-trained models show promising progress on various end tasks such as image retrieval and video captioning. Yet, they fail miserably on the recently proposed Winoground dataset, which challenges models to match paired images and English captions, with items constructed to overlap lexically but differ in meaning (e.g., "there is a mug in some grass" vs. "there is some grass in a mug"). By annotating the dataset using new fine-grained tags, we show that solving the Winoground task requires not just compositional language understanding, but a host of other abilities like commonsense reasoning or locating small, out-of-focus objects in low-resolution images. In this paper, we identify the dataset's main challenges through a suite of experiments on related tasks (probing task, image retrieval task), data augmentation, and manual inspection of the dataset. Our analysis suggests that a main challenge in visuolinguistic models may lie in fusing visual and textual representations, rather than in compositional language understanding. We release our annotation and code at https://github.com/ajd12342/why-winoground-hard .
This paper surveys vision-language pre-training (VLP) methods for multimodal intelligence that have been developed in the last few years. We group these approaches into three categories: ($i$) VLP for image-text tasks, such as image captioning, image-text retrieval, visual question answering, and visual grounding; ($ii$) VLP for core computer vision tasks, such as (open-set) image classification, object detection, and segmentation; and ($iii$) VLP for video-text tasks, such as video captioning, video-text retrieval, and video question answering. For each category, we present a comprehensive review of state-of-the-art methods, and discuss the progress that has been made and challenges still being faced, using specific systems and models as case studies. In addition, for each category, we discuss advanced topics being actively explored in the research community, such as big foundation models, unified modeling, in-context few-shot learning, knowledge, robustness, and computer vision in the wild, to name a few.
Searching troves of videos with textual descriptions is a core multimodal retrieval task. Owing to the lack of a purpose-built dataset for text-to-video retrieval, video captioning datasets have been re-purposed to evaluate models by (1) treating captions as positive matches to their respective videos and (2) assuming all other videos to be negatives. However, this methodology leads to a fundamental flaw during evaluation: since captions are marked as relevant only to their original video, many alternate videos also match the caption, which introduces false-negative caption-video pairs. We show that when these false negatives are corrected, a recent state-of-the-art model gains 25\% recall points -- a difference that threatens the validity of the benchmark itself. To diagnose and mitigate this issue, we annotate and release 683K additional caption-video pairs. Using these, we recompute effectiveness scores for three models on two standard benchmarks (MSR-VTT and MSVD). We find that (1) the recomputed metrics are up to 25\% recall points higher for the best models, (2) these benchmarks are nearing saturation for Recall@10, (3) caption length (generality) is related to the number of positives, and (4) annotation costs can be mitigated through sampling. We recommend retiring these benchmarks in their current form, and we make recommendations for future text-to-video retrieval benchmarks.
With the advent of rich visual representations and pre-trained language models, video captioning has seen continuous improvement over time. Despite the performance improvement, video captioning models are prone to hallucination. Hallucination refers to the generation of highly pathological descriptions that are detached from the source material. In video captioning, there are two kinds of hallucination: object and action hallucination. Instead of endeavoring to learn better representations of a video, in this work, we investigate the fundamental sources of the hallucination problem. We identify three main factors: (i) inadequate visual features extracted from pre-trained models, (ii) improper influences of source and target contexts during multi-modal fusion, and (iii) exposure bias in the training strategy. To alleviate these problems, we propose two robust solutions: (a) the introduction of auxiliary heads trained in multi-label settings on top of the extracted visual features and (b) the addition of context gates, which dynamically select the features during fusion. The standard evaluation metrics for video captioning measures similarity with ground truth captions and do not adequately capture object and action relevance. To this end, we propose a new metric, COAHA (caption object and action hallucination assessment), which assesses the degree of hallucination. Our method achieves state-of-the-art performance on the MSR-Video to Text (MSR-VTT) and the Microsoft Research Video Description Corpus (MSVD) datasets, especially by a massive margin in CIDEr score.
This paper tackles recipe generation from unsegmented cooking videos, a task that requires agents to (1) extract key events in completing the dish and (2) generate sentences for the extracted events. Our task is similar to dense video captioning (DVC), which aims at detecting events thoroughly and generating sentences for them. However, unlike DVC, in recipe generation, recipe story awareness is crucial, and a model should extract an appropriate number of events in the correct order and generate accurate sentences based on them. We analyze the output of the DVC model and confirm that although (1) several events are adoptable as a recipe story, (2) the generated sentences for such events are not grounded in the visual content. Based on this, we set our goal to obtain correct recipes by selecting oracle events from the output events and re-generating sentences for them. To achieve this, we propose a transformer-based multimodal recurrent approach of training an event selector and sentence generator for selecting oracle events from the DVC's events and generating sentences for them. In addition, we extend the model by including ingredients to generate more accurate recipes. The experimental results show that the proposed method outperforms state-of-the-art DVC models. We also confirm that, by modeling the recipe in a story-aware manner, the proposed model outputs the appropriate number of events in the correct order.
This paper presents OmniVL, a new foundation model to support both image-language and video-language tasks using one universal architecture. It adopts a unified transformer-based visual encoder for both image and video inputs, and thus can perform joint image-language and video-language pretraining. We demonstrate, for the first time, such a paradigm benefits both image and video tasks, as opposed to the conventional one-directional transfer (e.g., use image-language to help video-language). To this end, we propose a decoupled joint pretraining of image-language and video-language to effectively decompose the vision-language modeling into spatial and temporal dimensions and obtain performance boost on both image and video tasks. Moreover, we introduce a novel unified vision-language contrastive (UniVLC) loss to leverage image-text, video-text, image-label (e.g., image classification), video-label (e.g., video action recognition) data together, so that both supervised and noisily supervised pretraining data are utilized as much as possible. Without incurring extra task-specific adaptors, OmniVL can simultaneously support visual only tasks (e.g., image classification, video action recognition), cross-modal alignment tasks (e.g., image/video-text retrieval), and multi-modal understanding and generation tasks (e.g., image/video question answering, captioning). We evaluate OmniVL on a wide range of downstream tasks and achieve state-of-the-art or competitive results with similar model size and data scale.
Recent advances in text-to-image synthesis have led to large pretrained transformers with excellent capabilities to generate visualizations from a given text. However, these models are ill-suited for specialized tasks like story visualization, which requires an agent to produce a sequence of images given a corresponding sequence of captions, forming a narrative. Moreover, we find that the story visualization task fails to accommodate generalization to unseen plots and characters in new narratives. Hence, we first propose the task of story continuation, where the generated visual story is conditioned on a source image, allowing for better generalization to narratives with new characters. Then, we enhance or 'retro-fit' the pretrained text-to-image synthesis models with task-specific modules for (a) sequential image generation and (b) copying relevant elements from an initial frame. Then, we explore full-model finetuning, as well as prompt-based tuning for parameter-efficient adaptation, of the pre-trained model. We evaluate our approach StoryDALL-E on two existing datasets, PororoSV and FlintstonesSV, and introduce a new dataset DiDeMoSV collected from a video-captioning dataset. We also develop a model StoryGANc based on Generative Adversarial Networks (GAN) for story continuation, and compare it with the StoryDALL-E model to demonstrate the advantages of our approach. We show that our retro-fitting approach outperforms GAN-based models for story continuation and facilitates copying of visual elements from the source image, thereby improving continuity in the generated visual story. Finally, our analysis suggests that pretrained transformers struggle to comprehend narratives containing several characters. Overall, our work demonstrates that pretrained text-to-image synthesis models can be adapted for complex and low-resource tasks like story continuation.
Masked visual modeling (MVM) has been recently proven effective for visual pre-training. While similar reconstructive objectives on video inputs (e.g., masked frame modeling) have been explored in video-language (VidL) pre-training, previous studies fail to find a truly effective MVM strategy that can largely benefit the downstream performance. In this work, we systematically examine the potential of MVM in the context of VidL learning. Specifically, we base our study on a fully end-to-end VIdeO-LanguagE Transformer (VIOLET), where the supervision from MVM training can be backpropagated to the video pixel space. In total, eight different reconstructive targets of MVM are explored, from low-level pixel values and oriented gradients to high-level depth maps, optical flow, discrete visual tokens, and latent visual features. We conduct comprehensive experiments and provide insights into the factors leading to effective MVM training, resulting in an enhanced model VIOLETv2. Empirically, we show VIOLETv2 pre-trained with MVM objective achieves notable improvements on 13 VidL benchmarks, ranging from video question answering, video captioning, to text-to-video retrieval.
Current methods for text-to-video retrieval (T2VR) are trained and tested on video-captioning oriented datasets such as MSVD, MSR-VTT and VATEX. A key property of these datasets is that videos are assumed to be temporally pre-trimmed with short duration, whilst the provided captions well describe the gist of the video content. Consequently, for a given paired video and caption, the video is supposed to be fully relevant to the caption. In reality, however, as queries are not known a priori, pre-trimmed video clips may not contain sufficient content to fully meet the query. This suggests a gap between the literature and the real world. To fill the gap, we propose in this paper a novel T2VR subtask termed Partially Relevant Video Retrieval (PRVR). An untrimmed video is considered to be partially relevant w.r.t. a given textual query if it contains a moment relevant to the query. PRVR aims to retrieve such partially relevant videos from a large collection of untrimmed videos. PRVR differs from single video moment retrieval and video corpus moment retrieval, as the latter two are to retrieve moments rather than untrimmed videos. We formulate PRVR as a multiple instance learning (MIL) problem, where a video is simultaneously viewed as a bag of video clips and a bag of video frames. Clips and frames represent video content at different time scales. We propose a Multi-Scale Similarity Learning (MS-SL) network that jointly learns clip-scale and frame-scale similarities for PRVR. Extensive experiments on three datasets (TVR, ActivityNet Captions, and Charades-STA) demonstrate the viability of the proposed method. We also show that our method can be used for improving video corpus moment retrieval.
To generate proper captions for videos, the inference needs to identify relevant concepts and pay attention to the spatial relationships between them as well as to the temporal development in the clip. Our end-to-end encoder-decoder video captioning framework incorporates two transformer-based architectures, an adapted transformer for a single joint spatio-temporal video analysis as well as a self-attention-based decoder for advanced text generation. Furthermore, we introduce an adaptive frame selection scheme to reduce the number of required incoming frames while maintaining the relevant content when training both transformers. Additionally, we estimate semantic concepts relevant for video captioning by aggregating all ground truth captions of each sample. Our approach achieves state-of-the-art results on the MSVD, as well as on the large-scale MSR-VTT and the VATEX benchmark datasets considering multiple Natural Language Generation (NLG) metrics. Additional evaluations on diversity scores highlight the expressiveness and diversity in the structure of our generated captions.
Video-text retrieval (VTR) is an attractive yet challenging task for multi-modal understanding, which aims to search for relevant video (text) given a query (video). Existing methods typically employ completely heterogeneous visual-textual information to align video and text, whilst lacking the awareness of homogeneous high-level semantic information residing in both modalities. To fill this gap, in this work, we propose a novel visual-linguistic aligning model named HiSE for VTR, which improves the cross-modal representation by incorporating explicit high-level semantics. First, we explore the hierarchical property of explicit high-level semantics, and further decompose it into two levels, i.e. discrete semantics and holistic semantics. Specifically, for visual branch, we exploit an off-the-shelf semantic entity predictor to generate discrete high-level semantics. In parallel, a trained video captioning model is employed to output holistic high-level semantics. As for the textual modality, we parse the text into three parts including occurrence, action and entity. In particular, the occurrence corresponds to the holistic high-level semantics, meanwhile both action and entity represent the discrete ones. Then, different graph reasoning techniques are utilized to promote the interaction between holistic and discrete high-level semantics. Extensive experiments demonstrate that, with the aid of explicit high-level semantics, our method achieves the superior performance over state-of-the-art methods on three benchmark datasets, including MSR-VTT, MSVD and DiDeMo.