CVPR 2024 Papers

Large Language Models (LLMs) have been enhanced with vision capabilities enabling them to comprehend images videos and interleaved vision-language content. However the learning methods of these large multimodal models (LMMs) typically treat videos as predetermined clips rendering them less effective and efficient at handling streaming video inputs. In this paper we propose a novel Learning-In-Video-Stream (LIVE) framework which enables temporally aligned long-context and real-time dialogue within a continuous video stream. Our LIVE framework comprises comprehensive approaches to achieve video streaming dialogue encompassing: (1) a training objective designed to perform language modeling for continuous streaming inputs (2) a data generation scheme that converts offline temporal annotations into a streaming dialogue format and (3) an optimized inference pipeline to speed up interactive chat in real-world video streams. With our LIVE framework we develop a simplified model called VideoLLM-online and demonstrate its significant advantages in processing streaming videos. For instance our VideoLLM-online-7B model can operate at over 10 FPS on an A100 GPU for a 5-minute video clip from Ego4D narration. Moreover VideoLLM-online also showcases state-of-the-art performance on public offline video benchmarks such as recognition captioning and forecasting. The code model data and demo have been made available at showlab.github.io/videollm-online.

Video grounding aims to localize a spatio-temporal section in a video corresponding to an input text query. This paper addresses a critical limitation in current video grounding methodologies by introducing an Open-Vocabulary Spatio-Temporal Video Grounding task. Unlike prevalent closed-set approaches that struggle with open-vocabulary scenarios due to limited training data and predefined vocabularies our model leverages pre-trained representations from foundational spatial grounding models. This empowers it to effectively bridge the semantic gap between natural language and diverse visual content achieving strong performance in closed-set and open-vocabulary settings. Our contributions include a novel spatio-temporal video grounding model surpassing state-of-the-art results in closed-set evaluations on multiple datasets and demonstrating superior performance in open-vocabulary scenarios. Notably the proposed model outperforms state-of-the-art methods in closed-set settings on VidSTG (Declarative and Interrogative) and HC-STVG (V1 and V2) datasets. Furthermore in open-vocabulary evaluations on HC-STVG V1 and YouCook-Interactions our model surpasses the recent best-performing models by 4.88 m_vIoU and 1.83 accuracy demonstrating its efficacy in handling diverse linguistic and visual concepts for improved video understanding. Our codes will be publicly released.

Existing approaches to unsupervised video instance segmentation typically rely on motion estimates and experience difficulties tracking small or divergent motions. We present VideoCutLER a simple method for unsupervised multi-instance video segmentation without using motion-based learning signals like optical flow or training on natural videos. Our key insight is that using high-quality pseudo masks and a simple video synthesis method for model training is surprisingly sufficient to enable the resulting video model to effectively segment and track multiple instances across video frames. We show the first competitive unsupervised learning results on the challenging YouTubeVIS-2019 benchmark achieving 50.7% AP50 surpassing the previous state-of-the-art by a large margin. VideoCutLER can also serve as a strong pretrained model for supervised video instance segmentation tasks exceeding DINO by 15.9% on YouTubeVIS-2019 in terms of AP.

Text-to-video generation aims to produce a video based on a given prompt. Recently several commercial video models have been able to generate plausible videos with minimal noise excellent details and high aesthetic scores. However these models rely on large-scale well-filtered high-quality videos that are not accessible to the community. Many existing research works which train models using the low-quality WebVid-10M dataset struggle to generate high-quality videos because the models are optimized to fit WebVid-10M. In this work we explore the training scheme of video models extended from Stable Diffusion and investigate the feasibility of leveraging low-quality videos and synthesized high-quality images to obtain a high-quality video model. We first analyze the connection between the spatial and temporal modules of video models and the distribution shift to low-quality videos. We observe that full training of all modules results in a stronger coupling between spatial and temporal modules than only training temporal modules. Based on this stronger coupling we shift the distribution to higher quality without motion degradation by finetuning spatial modules with high-quality images resulting in a generic high-quality video model. Evaluations are conducted to demonstrate the superiority of the proposed method particularly in picture quality motion and concept composition.

Despite being (pre)trained on a massive amount of data state-of-the-art video-language alignment models are not robust to semantically-plausible contrastive changes in the video captions. Our work addresses this by identifying a broad spectrum of contrast misalignments such as replacing entities actions and flipping event order which alignment models should be robust against. To this end we introduce the VideoCon a video-language alignment dataset constructed by a large language model that generates plausible contrast video captions and explanations for differences between original and contrast video captions. Then a generative video-language model is finetuned with VideoCon to assess video-language entailment and generate explanations. Our VideoCon-based alignment model significantly outperforms current models. It exhibits a 12-point increase in AUC for the video-language alignment task on human-generated contrast captions. Finally our model sets new state of the art zero-shot performance in temporally-extensive video-language tasks such as text-to-video retrieval (SSv2-Temporal) and video question answering (ATP-Hard). Moreover our model shows superior performance on novel videos and human-crafted captions and explanations.

Text-driven video generation witnesses rapid progress. However merely using text prompts is not enough to depict the desired subject appearance that accurately aligns with users' intents especially for customized content creation. In this paper we study the task of video generation with image prompts which provide more accurate and direct content control beyond the text prompts. Specifically we propose a feed-forward framework VideoBooth with two dedicated designs: 1) We propose to embed image prompts in a coarse-to-fine manner. Coarse visual embeddings from image encoder provide high-level encodings of image prompts while fine visual embeddings from the proposed attention injection module provide multi-scale and detailed encoding of image prompts. These two complementary embeddings can faithfully capture the desired appearance. 2) In the attention injection module at fine level multi-scale image prompts are fed into different cross-frame attention layers as additional keys and values. This extra spatial information refines the details in the first frame and then it is propagated to the remaining frames which maintains temporal consistency. Extensive experiments demonstrate that VideoBooth achieves state-of-the-art performance in generating customized high-quality videos with subjects specified in image prompts. Notably VideoBooth is a generalizable framework where a single model works for a wide range of image prompts with only feed-forward passes.

Creating high-quality and interactive virtual environments such as games and simulators often involves complex and costly manual modeling processes. In this paper we present Video2Game a novel approach that automatically converts videos of real-world scenes into realistic and interactive game environments. At the heart of our system are three core components: (i) a neural radiance fields (NeRF) module that effectively captures the geometry and visual appearance of the scene; (ii) a mesh module that distills the knowledge from NeRF for faster rendering; and (iii) a physics module that models the interactions and physical dynamics among the objects. By following the carefully designed pipeline one can construct an interactable and actionable digital replica of the real world. We benchmark our system on both indoor and large-scale outdoor scenes. We show that we can not only produce highly-realistic renderings in real-time but also build interactive games on top.

Video-P2P is the first framework for real-world video editing with cross-attention control. While attention control has proven effective for image editing with pre-trained image generation models there are currently no large-scale video generation models publicly available. Video-P2P addresses this limitation by adapting an image generation diffusion model to complete various video editing tasks. Specifically we propose to first tune a Text-to-Set (T2S) model to complete an approximate inversion and then optimize a shared unconditional embedding to achieve accurate video inversion with a small memory cost. We further prove that it is crucial for consistent video editing. For attention control we introduce a novel decoupled-guidance strategy which uses different guidance strategies for the source and target prompts. The optimized unconditional embedding for the source prompt improves reconstruction ability while an initialized unconditional embedding for the target prompt enhances editability. Incorporating the attention maps of these two branches enables detailed editing. These technical designs enable various text-driven editing applications including word swap prompt refinement and attention re-weighting. Video-P2P works well on real-world videos for generating new characters while optimally preserving their original poses and scenes. It significantly outperforms previous approaches.

By leveraging temporal dependency in video sequences multi-frame human pose estimation algorithms have demonstrated remarkable results in complicated situations such as occlusion motion blur and video defocus. These algorithms are predominantly based on heatmaps resulting in high computation and storage requirements per frame which limits their flexibility and real-time application in video scenarios particularly on edge devices. In this paper we develop an efficient and effective video-based human pose regression method which bypasses intermediate representations such as heatmaps and instead directly maps the input to the output joint coordinates. Despite the inherent spatial correlation among adjacent joints of the human pose the temporal trajectory of each individual joint exhibits relative independence. In light of this we propose a novel Decoupled Space-Time Aggregation network (DSTA) to separately capture the spatial contexts between adjacent joints and the temporal cues of each individual joint thereby avoiding the conflation of spatiotemporal dimensions. Concretely DSTA learns a dedicated feature token for each joint to facilitate the modeling of their spatiotemporal dependencies. With the proposed joint-wise local-awareness attention mechanism our method is capable of efficiently and flexibly utilizing the spatial dependency of adjacent joints and the temporal dependency of each joint itself. Extensive experiments demonstrate the superiority of our method. Compared to previous regression-based single-frame human pose estimation methods DSTA significantly enhances performance achieving an 8.9 mAP improvement on PoseTrack2017. Furthermore our approach either surpasses or is on par with the state-of-the-art heatmap-based multi-frame human pose estimation methods. Project page: https://github.com/zgspose/DSTA.

The creation of new datasets often presents new challenges for video recognition and can inspire novel ideas while addressing these challenges. While existing datasets mainly comprise landscape mode videos our paper seeks to introduce portrait mode videos to the research community and highlight the unique challenges associated with this video format. With the growing popularity of smartphones and social media applications recognizing portrait mode videos is becoming increasingly important. To this end we have developed the first dataset dedicated to portrait mode video recognition namely PortraitMode-400. The taxonomy of PortraitMode-400 was constructed in a data-driven manner comprising 400 fine-grained categories and rigorous quality assurance was implemented to ensure the accuracy of human annotations. In addition to the new dataset we conducted a comprehensive analysis of the impact of video format (portrait mode versus landscape mode) on recognition accuracy and spatial bias due to the different formats. Furthermore we designed extensive experiments to explore key aspects of portrait mode video recognition including the choice of data augmentation evaluation procedure the importance of temporal information and the role of audio modality. Building on the insights from our experimental results and the introduction of PortraitMode-400 our paper aims to inspire further research efforts in this emerging research area.

Most video captioning models are designed to process short video clips of few seconds and output text describing low-level visual concepts (e.g. objects scenes atomic actions). However most real-world videos last for minutes or hours and have a complex hierarchical structure spanning different temporal granularities. We propose Video ReCap a recursive video captioning model that can process video inputs of dramatically different lengths (from 1 second to 2 hours) and output video captions at multiple hierarchy levels. The recursive video-language architecture exploits the synergy between different video hierarchies and can process hour-long videos efficiently. We utilize a curriculum learning training scheme to learn the hierarchical structure of videos starting from clip-level captions describing atomic actions then focusing on segment-level descriptions and concluding with generating summaries for hour-long videos. Furthermore we introduce Ego4D-HCap dataset by augmenting Ego4D with 8267 manually collected long-range video summaries. Our recursive model can flexibly generate captions at different hierarchy levels while also being useful for other complex video understanding tasks such as VideoQA on EgoSchema. Data code and models are publicly available at https://sites.google.com/view/vidrecap.

Diffusion models have made significant strides in image generation mastering tasks such as unconditional image synthesis text-image translation and image-to-image conversions. However their capability falls short in the realm of video prediction mainly because they treat videos as a collection of independent images relying on external constraints such as temporal attention mechanisms to enforce temporal coherence. In our paper we introduce a novel model class that treats video as a continuous multi-dimensional process rather than a series of discrete frames. Through extensive experimentation we establish state-of-the-art performance in video prediction validated on benchmark datasets including KTH BAIR Human3.6M and UCF101.

We present VIDIM a generative model for video interpolation which creates short videos given a start and end frame. In order to achieve high fidelity and generate motions unseen in the input data VIDIM uses cascaded diffusion models to first generate the target video at low resolution and then generate the high-resolution video conditioned on the low-resolution generated video. We compare VIDIM to previous state-of-the-art methods on video interpolation and demonstrate how such works fail in most settings where the underlying motion is complex nonlinear or ambiguous while VIDIM can easily handle such cases. We additionally demonstrate how classifier-free guidance on the start and end frame and conditioning the superresolution model on the original high-resolution frames without additional parameters unlocks high-fidelity results. VIDIM is fast to sample from as it jointly denoises all the frames to be generated requires less than a billion parameters per diffusion model to produce compelling results and still enjoys scalability and improved quality at larger parameter counts. Please see our project page at vidiminterpolation.github.io.

Video harmonization is an important and challenging task that aims to obtain visually realistic composite videos by automatically adjusting the foreground's appearance to harmonize with the background. Inspired by the short-term and long-term gradual adjustment process of manual harmonization we present a Video Triplet Transformer framework to model three spatio-temporal variation patterns within videos i.e. short-term spatial as well as long-term global and dynamic for video-to-video tasks like video harmonization. Specifically for short-term harmonization we adjust foreground appearance to consist with background in spatial dimension based on the neighbor frames; for long-term harmonization we not only explore global appearance variations to enhance temporal consistency but also alleviate motion offset constraints to align similar contextual appearances dynamically. Extensive experiments and ablation studies demonstrate the effectiveness of our method achieving state-of-the-art performance in video harmonization video enhancement and video demoireing tasks. We also propose a temporal consistency metric to better evaluate the harmonized videos. Code is available at https://github.com/zhenglab/VideoTripletTransformer.

Video Frame Interpolation (VFI) has witnessed a surge in popularity due to its abundant downstream applications. Event-based VFI (E-VFI) has recently propelled the advancement of VFI. Thanks to the high temporal resolution benefits event cameras can bridge the informational void present between successive video frames. Most state-of-the-art E-VFI methodologies follow the conventional VFI paradigm which pivots on motion estimation between consecutive frames to generate intermediate frames through a process of warping and refinement. However this reliance engenders a heavy dependency on the quality and consistency of keyframes rendering these methods susceptible to challenges in extreme real-world scenarios such as missing moving objects and severe occlusion dilemmas. This study proposes a novel E-VFI framework that directly synthesize intermediate frames leveraging event-based reference obviating the necessity for explicit motion estimation and substantially enhancing the capacity to handle motion occlusion. Given the sparse and inherently noisy nature of event data we prioritize the reliability of the event-based reference leading to the development of an innovative event-aware reconstruction strategy for accurate reference generation. Besides we implement a bi-directional event-guided alignment from keyframes to the reference using the introduced E-PCD module. Finally a transformer-based decoder is adopted for prediction refinement. Comprehensive experimental evaluations on both synthetic and real-world datasets underscore the superiority of our approach and its potential to execute high-quality VFI tasks.

Diffusion models have made significant advances in generating high-quality images but their application to video generation has remained challenging due to the complexity of temporal motion. Zero-shot video editing offers a solution by utilizing pre-trained image diffusion models to translate source videos into new ones. Nevertheless existing methods struggle to maintain strict temporal consistency and efficient memory consumption. In this work we propose a novel approach to enhance temporal consistency in generated videos by merging self-attention tokens across frames. By aligning and compressing temporally redundant tokens across frames our method improves temporal coherence and reduces memory consumption in self-attention computations. The merging strategy matches and aligns tokens according to the temporal correspondence between frames facilitating natural temporal consistency in generated video frames. To manage the complexity of video processing we divide videos into chunks and develop intra-chunk local token merging and inter-chunk global token merging ensuring both short-term video continuity and long-term content consistency. Our video editing approach seamlessly extends the advancements in image editing to video editing rendering favorable results in temporal consistency over state-of-the-art methods.

In this paper we propose VidLA an approach for video-language alignment at scale. There are two major limitations of previous video-language alignment approaches. First they do not capture both short-range and long-range temporal dependencies and typically employ complex hierarchical deep network architectures that are hard to integrate with existing pretrained image-text foundation models. To effectively address this limitation we instead keep the network architecture simple and use a set of data tokens that operate at different temporal resolutions in a hierarchical manner accounting for the temporally hierarchical nature of videos. By employing a simple two-tower architecture we are able to initialize our video-language model with pretrained image-text foundation models thereby boosting the final performance. Second existing video-language alignment works struggle due to the lack of semantically aligned large-scale training data. To overcome it we leverage recent LLMs to curate the largest video-language dataset to date with better visual grounding. Furthermore unlike existing video-text datasets which only contain short clips our dataset is enriched with video clips of varying durations to aid our temporally hierarchical data tokens in extracting better representations at varying temporal scales. Overall empirical results show that our proposed approach surpasses state-of-the-art methods on multiple retrieval benchmarks especially on longer videos and performs competitively on classification benchmarks.

Vision-Language models (VLMs) have excelled in the image-domain--- especially in zero-shot settings--- thanks to the availability of vast pretraining data (i.e. paired image-text samples). However for videos such paired data is not as abundant. Therefore video-VLMs are usually designed by adapting pretrained image-VLMs to the video-domain instead of training from scratch. All such recipes rely on augmenting visual embeddings with temporal information (i.e. image --> video) often keeping text embeddings unchanged or even being discarded. In this paper we argue the contrary that better video-VLMs can be designed by focusing more on augmenting text rather than visual information. More specifically we introduce Video-conditioned Text Representations (VicTR): a form of text embeddings optimized w.r.t. visual embeddings creating a more-flexible contrastive latent space. Our model can further make use of freely-available semantic information in the form of visually-grounded auxiliary text (e.g. object or scene information). We evaluate our model on few-shot zero-shot (HMDB-51 UCF-101) short-form (Kinetics-400) and long-form (Charades) activity recognition benchmarks showing strong performance among video-VLMs.