Factorized Latent Dynamics for Video JEPA: An Empirical Study of Auxiliary Objectives

• URL: http://arxiv.org/abs/2605.17165v1
• Date: Sat, 16 May 2026 21:30:51 GMT
• Title: Factorized Latent Dynamics for Video JEPA: An Empirical Study of Auxiliary Objectives
• Abstract summary: Joint-Embedding Predictive Architectures (JEPA) are a promising framework for self-supervised video representation learning.<n>We report a small-scale empirical study of 18 auxiliary objective variants for Video-JEPA across two pretraining regimes.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Joint-Embedding Predictive Architectures (JEPA) are a promising framework for self-supervised video representation learning, yet the behavior of auxiliary objectives in small-scale Video-JEPA training is not well characterized. We report a small-scale empirical study of 18 auxiliary objective variants for Video-JEPA across two pretraining regimes: single-dataset (UCF-101) and mixed-dataset (UCF-101 + Something-Something V2 + ImageNet-100). We evaluate frozen representations on three complementary benchmarks: Diving-48 (fine-grained motion), SomethingSomething V2 (temporal reasoning), and ImageNet-100 (appearance). Our experiments suggest that many auxiliary objectives exhibit capacity trade-offs: gains on one downstream capability often coincide with degradation on another. We then study FWM-HW-LD (Factorized World-Model with Hard-Region-Weighted Latent Dynamics), a training-time objective that separates the latent representation into appearance and dynamics subspaces and applies hard-region weighting to both JEPA prediction errors and latent dynamics errors. In our mixed-dataset setting, FWM-HW-LD improves ImageNet-100 by +5.92 and SSv2 by +3.21 percentage points relative to the reference baseline, while remaining within 0.30 percentage points on Diving-48. These results indicate that latent factorization is a useful direction for studying auxiliary-objective trade-offs in Video-JEPA.

Related papers

• V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learning [46.89986408533846]
  V-JEPA 2.1 is a family of self-supervised models that learn dense, high-quality visual representations for both images and videos.<n>We show that V-JEPA 2.1 significantly advances the state-of-the-art in dense visual understanding and world modeling.
  arXiv  Detail & Related papers  (2026-03-15T17:02:40Z)
• A Lightweight Library for Energy-Based Joint-Embedding Predictive Architectures [58.26804959656713]
  We present EB-JEPA, an open-source library for learning representations and world models using Joint-Embedding Predictive Architectures (JEPAs)<n>JEPAs learn to predict in representation space rather than pixel space, avoiding the pitfalls of generative modeling.<n>We show how these representations can drive action-conditioned world models, achieving a 97% planning success rate on the Two Rooms navigation task.
  arXiv  Detail & Related papers  (2026-02-03T14:56:24Z)
• LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics [53.247652209132376]
  Joint-Embedding Predictive Architectures (JEPAs) offer a promising blueprint, but lack of practical guidance and theory has led to ad-hoc R&D.<n>We present a comprehensive theory of JEPAs and instantiate it in bf LeJEPA, a lean, scalable, and theoretically grounded training objective.
  arXiv  Detail & Related papers  (2025-11-11T18:21:55Z)
• Temporal vs. Spatial: Comparing DINOv3 and V-JEPA2 Feature Representations for Video Action Analysis [0.0]
  DINOv3 processes frames independently through spatial feature extraction, and V-JEPA2 employs joint temporal modeling across video sequences.<n>We evaluate both approaches on the UCF Sports dataset, examining feature quality through multiple dimensions including classification accuracy, clustering performance, intra-class consistency, and inter-class discrimination.
  arXiv  Detail & Related papers  (2025-09-25T21:05:07Z)
• MoE-LLaVA: Mixture of Experts for Large Vision-Language Models [49.32669226551026]
  We propose a simple yet effective training strategy MoE-Tuning for LVLMs.<n>MoE-LLaVA, a MoE-based sparse LVLM architecture, uniquely activates only the top-k experts through routers.<n>Experiments show the significant performance of MoE-LLaVA in a variety of visual understanding and object hallucination benchmarks.
  arXiv  Detail & Related papers  (2024-01-29T08:13:40Z)
• When hard negative sampling meets supervised contrastive learning [17.173114048398947]
  We introduce a new supervised contrastive learning objective, SCHaNe, which incorporates hard negative sampling during the fine-tuning phase. SchaNe outperforms the strong baseline BEiT-3 in Top-1 accuracy across various benchmarks. Our proposed objective sets a new state-of-the-art for base models on ImageNet-1k, achieving an 86.14% accuracy.
  arXiv  Detail & Related papers  (2023-08-28T20:30:10Z)
• Class Anchor Margin Loss for Content-Based Image Retrieval [97.81742911657497]
  We propose a novel repeller-attractor loss that falls in the metric learning paradigm, yet directly optimize for the L2 metric without the need of generating pairs. We evaluate the proposed objective in the context of few-shot and full-set training on the CBIR task, by using both convolutional and transformer architectures.
  arXiv  Detail & Related papers  (2023-06-01T12:53:10Z)
• Motion Sensitive Contrastive Learning for Self-supervised Video Representation [34.854431881562576]
  Motion Sensitive Contrastive Learning (MSCL) injects the motion information captured by optical flows into RGB frames to strengthen feature learning. Local Motion Contrastive Learning (LMCL) with frame-level contrastive objectives across the two modalities. Flow Rotation Augmentation (FRA) to generate extra motion-shuffled negative samples and Motion Differential Sampling (MDS) to accurately screen training samples.
  arXiv  Detail & Related papers  (2022-08-12T04:06:56Z)
• Anchor Retouching via Model Interaction for Robust Object Detection in Aerial Images [15.404024559652534]
  We present an effective Dynamic Enhancement Anchor (DEA) network to construct a novel training sample generator. Our method achieves state-of-the-art performance in accuracy with moderate inference speed and computational overhead for training.
  arXiv  Detail & Related papers  (2021-12-13T14:37:20Z)
• Disentangle Your Dense Object Detector [82.22771433419727]
  Deep learning-based dense object detectors have achieved great success in the past few years and have been applied to numerous multimedia applications such as video understanding. However, the current training pipeline for dense detectors is compromised to lots of conjunctions that may not hold. We propose Disentangled Dense Object Detector (DDOD), in which simple and effective disentanglement mechanisms are designed and integrated into the current state-of-the-art detectors.
  arXiv  Detail & Related papers  (2021-07-07T00:52:16Z)

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.