Long-Context Autoregressive Video Modeling with Next-Frame Prediction

• URL: http://arxiv.org/abs/2503.19325v3
• Date: Sun, 18 May 2025 02:27:31 GMT
• Title: Long-Context Autoregressive Video Modeling with Next-Frame Prediction
• Authors: Yuchao Gu, Weijia Mao, Mike Zheng Shou,
• Abstract summary: Long-context video modeling is essential for enabling generative models to function as world simulators.<n>While training directly on long videos is a natural solution, the rapid growth of vision tokens makes it computationally prohibitive.<n>We propose Frame AutoRegressive (FAR) models temporal dependencies between continuous frames, converges faster than video diffusion transformers, and outperforms token-level autoregressive models.
• Score: 17.710915002557996
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Long-context video modeling is essential for enabling generative models to function as world simulators, as they must maintain temporal coherence over extended time spans. However, most existing models are trained on short clips, limiting their ability to capture long-range dependencies, even with test-time extrapolation. While training directly on long videos is a natural solution, the rapid growth of vision tokens makes it computationally prohibitive. To support exploring efficient long-context video modeling, we first establish a strong autoregressive baseline called Frame AutoRegressive (FAR). FAR models temporal dependencies between continuous frames, converges faster than video diffusion transformers, and outperforms token-level autoregressive models. Based on this baseline, we observe context redundancy in video autoregression. Nearby frames are critical for maintaining temporal consistency, whereas distant frames primarily serve as context memory. To eliminate this redundancy, we propose the long short-term context modeling using asymmetric patchify kernels, which apply large kernels to distant frames to reduce redundant tokens, and standard kernels to local frames to preserve fine-grained detail. This significantly reduces the training cost of long videos. Our method achieves state-of-the-art results on both short and long video generation, providing an effective baseline for long-context autoregressive video modeling.

Related papers

• LongDWM: Cross-Granularity Distillation for Building a Long-Term Driving World Model [22.92353994818742]
  Driving world models are used to simulate futures by video generation based on the condition of the current state and actions.<n>Recent studies utilize the Diffusion Transformer (DiT) as the backbone of driving world models to improve learning flexibility.<n>We propose several solutions to build a simple yet effective long-term driving world model.
  arXiv  Detail & Related papers  (2025-06-02T11:19:23Z)
• Multimodal Long Video Modeling Based on Temporal Dynamic Context [13.979661295432964]
  We propose a dynamic long video encoding method utilizing the temporal relationship between frames, named Temporal Dynamic Context (TDC) We segment the video into semantically consistent scenes based on inter-frame similarities, then encode each frame into tokens using visual-audio encoders. To handle extremely long videos, we propose a training-free chain-of-thought strategy that progressively extracts answers from multiple video segments.
  arXiv  Detail & Related papers  (2025-04-14T17:34:06Z)
• BIMBA: Selective-Scan Compression for Long-Range Video Question Answering [46.199493246921435]
  Video Question Answering (VQA) in long videos poses the key challenge of extracting relevant information. We introduce BIMBA, an efficient state-space model to handle long-form videos.
  arXiv  Detail & Related papers  (2025-03-12T17:57:32Z)
• Token-Efficient Long Video Understanding for Multimodal LLMs [101.70681093383365]
  STORM is a novel architecture incorporating a dedicated temporal encoder between the image encoder and the Video-LLMs.<n>We show that STORM achieves state-of-the-art results across various long video understanding benchmarks.
  arXiv  Detail & Related papers  (2025-03-06T06:17:38Z)
• LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding [65.46303012350207]
  LongVU is an adaptive compression mechanism that reduces the number of video tokens while preserving visual details of long videos. We leverage DINOv2 features to remove redundant frames that exhibit high similarity. We perform spatial token reduction across frames based on their temporal dependencies.
  arXiv  Detail & Related papers  (2024-10-22T21:21:37Z)
• Long Context Transfer from Language to Vision [74.78422371545716]
  Video sequences offer valuable temporal information, but existing large multimodal models (LMMs) fall short in understanding extremely long videos. In this paper, we approach this problem from the perspective of the language model. By simply extrapolating the context length of the language backbone, we enable LMMs to comprehend orders of magnitude more visual tokens without any video training.
  arXiv  Detail & Related papers  (2024-06-24T17:58:06Z)
• ViD-GPT: Introducing GPT-style Autoregressive Generation in Video Diffusion Models [66.84478240757038]
  A majority of video diffusion models (VDMs) generate long videos in an autoregressive manner, i.e., generating subsequent clips conditioned on last frames of previous clip. We introduce causal (i.e., unidirectional) generation into VDMs, and use past frames as prompt to generate future frames. Our ViD-GPT achieves state-of-the-art performance both quantitatively and qualitatively on long video generation.
  arXiv  Detail & Related papers  (2024-06-16T15:37:22Z)
• ZeroSmooth: Training-free Diffuser Adaptation for High Frame Rate Video Generation [81.90265212988844]
  We propose a training-free video method for generative video models in a plug-and-play manner. We transform a video model into a self-cascaded video diffusion model with the designed hidden state correction modules. Our training-free method is even comparable to trained models supported by huge compute resources and large-scale datasets.
  arXiv  Detail & Related papers  (2024-06-03T00:31:13Z)
• Mirasol3B: A Multimodal Autoregressive model for time-aligned and contextual modalities [67.89368528234394]
  One of the main challenges of multimodal learning is the need to combine heterogeneous modalities. Video and audio are obtained at much higher rates than text and are roughly aligned in time. Our approach achieves the state-of-the-art on well established multimodal benchmarks, outperforming much larger models.
  arXiv  Detail & Related papers  (2023-11-09T19:15:12Z)
• Long-Form Video-Language Pre-Training with Multimodal Temporal Contrastive Learning [39.80936685227549]
  Large-scale video-language pre-training has shown significant improvement in video-language understanding tasks. We introduce a Long-Form VIdeo-LAnguage pre-training model (VILA) and train it on a large-scale long-form video and paragraph dataset. We fine-tune the model on seven downstream long-form video-language understanding tasks, achieve new state-of-the-art performances.
  arXiv  Detail & Related papers  (2022-10-12T09:08:27Z)
• Learning Fine-Grained Visual Understanding for Video Question Answering via Decoupling Spatial-Temporal Modeling [28.530765643908083]
  We decouple spatial-temporal modeling and integrate an image- and a video-language to learn fine-grained visual understanding. We propose a novel pre-training objective, Temporal Referring Modeling, which requires the model to identify temporal positions of events in video sequences. Our model outperforms previous work pre-trained on orders of magnitude larger datasets.
  arXiv  Detail & Related papers  (2022-10-08T07:03:31Z)
• Generating Long Videos of Dynamic Scenes [66.56925105992472]
  We present a video generation model that reproduces object motion, changes in camera viewpoint, and new content that arises over time. A common failure case is for content to never change due to over-reliance on inductive biases to provide temporal consistency.
  arXiv  Detail & Related papers  (2022-06-07T16:29:51Z)
• Learning Trajectory-Aware Transformer for Video Super-Resolution [50.49396123016185]
  Video super-resolution aims to restore a sequence of high-resolution (HR) frames from their low-resolution (LR) counterparts. Existing approaches usually align and aggregate video frames from limited adjacent frames. We propose a novel Transformer for Video Super-Resolution (TTVSR)
  arXiv  Detail & Related papers  (2022-04-08T03:37:39Z)
• Video Demoireing with Relation-Based Temporal Consistency [68.20281109859998]
  Moire patterns, appearing as color distortions, severely degrade image and video qualities when filming a screen with digital cameras. We study how to remove such undesirable moire patterns in videos, namely video demoireing.
  arXiv  Detail & Related papers  (2022-04-06T17:45:38Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.