Representations Before Pixels: Semantics-Guided Hierarchical Video Prediction

• URL: http://arxiv.org/abs/2604.11707v1
• Date: Mon, 13 Apr 2026 16:42:46 GMT
• Title: Representations Before Pixels: Semantics-Guided Hierarchical Video Prediction
• Abstract summary: We present Re2Pix, a hierarchical video prediction framework.<n>It decomposes forecasting into two stages: semantic representation prediction and representation-guided visual synthesis.<n>A key challenge arises from the train-test mismatch between ground-truth representations available during training and predicted ones used at inference.
• Score: 12.661995278049764
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Accurate future video prediction requires both high visual fidelity and consistent scene semantics, particularly in complex dynamic environments such as autonomous driving. We present Re2Pix, a hierarchical video prediction framework that decomposes forecasting into two stages: semantic representation prediction and representation-guided visual synthesis. Instead of directly predicting future RGB frames, our approach first forecasts future scene structure in the feature space of a frozen vision foundation model, and then conditions a latent diffusion model on these predicted representations to render photorealistic frames. This decomposition enables the model to focus first on scene dynamics and then on appearance generation. A key challenge arises from the train-test mismatch between ground-truth representations available during training and predicted ones used at inference. To address this, we introduce two conditioning strategies, nested dropout and mixed supervision, that improve robustness to imperfect autoregressive predictions. Experiments on challenging driving benchmarks demonstrate that the proposed semantics-first design significantly improves temporal semantic consistency, perceptual quality, and training efficiency compared to strong diffusion baselines. We provide the implementation code at https://github.com/Sta8is/Re2Pix

Related papers

• Native Video-Action Pretraining for Generalizable Robot Control [77.59831077077176]
  LingBot-VA 2.0 is a video-action foundation model built from the ground up for embodiment.<n>To meet the demands of high-frequency inference, our model employs a sparse MoE backbone.<n>Real-world deployment validates LingBot-VA 2.0 as a robust foundation model.
  arXiv  Detail & Related papers  (2026-07-09T16:15:43Z)
• FLaRA: Predicting Future Latent Representations for Accident Anticipation [15.994549120609358]
  Anticipating traffic accidents from dashcam videos is a critical challenge in intelligent transportation systems.<n>We propose FLaRA, a novel predictive architecture that shifts this paradigm by forecasting future latent representations for accident anticipation.<n>Our approach achieves state-of-the-art performance while maintaining realistic early warning capabilities.
  arXiv  Detail & Related papers  (2026-06-12T12:16:45Z)
• ForecastOcc: Vision-based Semantic Occupancy Forecasting [16.699381591572163]
  We present ForecastOcc, the first framework for vision-based semantic occupancy forecasting that predicts future occupancy states and semantic categories.<n>Our framework yields semantic occupancy forecasts for multiple horizons directly from past camera images, without relying on externally estimated maps.
  arXiv  Detail & Related papers  (2026-02-08T15:16:06Z)
• Learning from Next-Frame Prediction: Autoregressive Video Modeling Encodes Effective Representations [53.91818843831925]
  We propose NExT-Vid, a novel autoregressive visual generative pretraining framework.<n>We introduce a context-isolated autoregressive predictor to decouple semantic representation from target decoding.<n>Through context-isolated flow-matching pretraining, our approach achieves strong representations.
  arXiv  Detail & Related papers  (2025-12-24T07:07:08Z)
• GaussianPrediction: Dynamic 3D Gaussian Prediction for Motion Extrapolation and Free View Synthesis [71.24791230358065]
  We introduce a novel framework that empowers 3D Gaussian representations with dynamic scene modeling and future scenario synthesis. GaussianPrediction can forecast future states from any viewpoint, using video observations of dynamic scenes. Our framework shows outstanding performance on both synthetic and real-world datasets, demonstrating its efficacy in predicting and rendering future environments.
  arXiv  Detail & Related papers  (2024-05-30T06:47:55Z)
• State-space Decomposition Model for Video Prediction Considering Long-term Motion Trend [3.910356300831074]
  We propose a state-space decomposition video prediction model that decomposes the overall video frame generation into deterministic appearance prediction and motion prediction. We infer the long-term motion trend from conditional frames to guide the generation of future frames that exhibit high consistency with the conditional frames.
  arXiv  Detail & Related papers  (2024-04-17T17:19:48Z)
• Predicting Long-horizon Futures by Conditioning on Geometry and Time [49.86180975196375]
  We explore the task of generating future sensor observations conditioned on the past. We leverage the large-scale pretraining of image diffusion models which can handle multi-modality. We create a benchmark for video prediction on a diverse set of videos spanning indoor and outdoor scenes.
  arXiv  Detail & Related papers  (2024-04-17T16:56:31Z)
• LOPR: Latent Occupancy PRediction using Generative Models [28.49346874213506]
  LiDAR generated occupancy grid maps (L-OGMs) offer a robust bird's eye-view scene representation.<n>We propose a framework that decouples occupancy prediction into: representation learning and prediction within the learned latent space.
  arXiv  Detail & Related papers  (2022-10-03T22:04:00Z)
• HARP: Autoregressive Latent Video Prediction with High-Fidelity Image Generator [90.74663948713615]
  We train an autoregressive latent video prediction model capable of predicting high-fidelity future frames. We produce high-resolution (256x256) videos with minimal modification to existing models.
  arXiv  Detail & Related papers  (2022-09-15T08:41:57Z)
• Fourier-based Video Prediction through Relational Object Motion [28.502280038100167]
  Deep recurrent architectures have been applied to the task of video prediction. Here, we explore a different approach by using frequency-domain approaches for video prediction. The resulting predictions are consistent with the observed dynamics in a scene and do not suffer from blur.
  arXiv  Detail & Related papers  (2021-10-12T10:43:05Z)
• Learning Semantic-Aware Dynamics for Video Prediction [68.04359321855702]
  We propose an architecture and training scheme to predict video frames by explicitly modeling dis-occlusions. The appearance of the scene is warped from past frames using the predicted motion in co-visible regions.
  arXiv  Detail & Related papers  (2021-04-20T05:00:24Z)

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.