SV-TAD introduces native sparse 2D convolutions to enable convolutional adapters to operate directly on dynamically pruned token sets, addressing the inefficiency of dense reconstruction in token‑selection methods. By integrating this primitive into a temporal action detection framework, the approach cuts VideoMAEv2‑L computation by up to 64% and achieves 2.2× faster inference while preserving state‑of‑the‑art accuracy on THUMOS‑14 and ActivityNet‑1.3. When applied to InternVideoNext‑L, it surpasses prior state‑of‑the‑art performance at roughly half the computational cost and naturally supports auxiliary task tokens for improved fine‑grained assembly detection on ATTACH.
To adapt billion-parameter Vision Transformers for long-video understanding, recent methods freeze the backbone and train lightweight convolutional modules. While effective for parameter-efficient tra...
arXiv:2609.38984v1 Announce Type: cross Abstract: World-action models (WAMs) leverage pretrained video models to improve generalization in robot control by jointly predicting future visual states and...
arXiv:2606. 20246v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models pre-trained on massive video-robot datasets have revolutionized robotic manipulation, yet their multi-billion parameter architectures impose prohibitive computational burdens during downstream fine-tuning and real-time inference.
DynamicVLA is a latency‑aware Vision‑Language‑Action model designed for dynamic object manipulation, featuring a compact 0.4B architecture and a convolutional vision encoder for efficient multimodal inference. It employs a continuous inference schedule that overlaps reasoning and execution, and a Latent‑aware Action Streaming mechanism that discards stale action prefixes to maintain action‑time alignment. The authors also introduce the Dynamic Object Manipulation (DOM) benchmark, comprising 200K synthetic episodes and 2K real‑world episodes, and demonstrate that DynamicVLA improves dynamic manipulation success in simulation and on real robots.
VideoTIR introduces a reinforcement‑learning approach to improve long‑video understanding by encouraging multimodal large language models to use comprehensive multi‑level toolkits efficiently. It combines Zero‑RL and SFT cold‑starting strategies to help models retrieve and focus on meaningful video segments, images, and regions, thereby reducing hallucinations. The method includes Toolkit Action Grouped Policy Optimization (TAGPO) to streamline tool‑calling and a sandbox‑based trajectory synthesis framework for high‑quality data, achieving strong results on three long‑video QA benchmarks.
VETO (Video Efficient Token Optimization for Vision Language Models) is a plug‑in that reduces the quadratic cost of visual tokens in long‑video inference by applying dual‑axis compression: an intra‑frame compressor merges semantically similar tokens within each frame, and an inter‑frame compressor merges temporally redundant frames. By first compressing spatial dimensions, VETO lowers the cost of subsequent global temporal matching, surpassing single‑axis methods and achieving up to 45% faster inference on models such as LLaVA‑OneVision‑7B while maintaining or improving accuracy. The approach is universally applicable across LLaVA‑OneVision, InternVL‑2.5, and LongVA, preserving or enhancing zero‑shot accuracy even under extreme token budgets.
The paper introduces Skeleton-Language feature Pooling Switching, a weakly‑supervised vision‑language pretraining strategy for skeleton‑based zero‑shot spatio‑temporal action localization. It replaces video‑level pooling with instance‑level feature computation during inference, enabling the model to estimate unseen actions without costly annotations. Additionally, Scene‑Mixed Discriminative Contrastive Learning is proposed to separate actions at the instance level within mixed scenes using a MIL framework, and experiments on four public datasets confirm the method’s effectiveness.
arXiv:2607. 13421v1 Announce Type: cross Abstract: Spatio-Temporal Video Grounding (STVG) aims to retrieve the visual trajectory of a specific object from a video stream as described by a natural language expression.
arXiv:2507. 05116v5 Announce Type: replace-cross Abstract: Recent large-scale Vision Language Action (VLA) models have shown superior performance in robotic manipulation tasks guided by natural language.
arXiv:2607.14935v2 Announce Type: replace Abstract: Recent advances in video understanding have spanned motion, long video, and streaming interaction, driving this field toward real-world application...
WALL-WM is a World Action Model that shifts video-action learning from chunk-centric optimization to event-grounded Vision-Language-Action pretraining, using semantically coherent action events as the atomic unit of learning. Existing WAMs commonly initialize from multimodal or video foundation models and then optimize fixed-length action chunks conditioned directly on the current observation and instruction.
The paper introduces STITCH, a training‑free method that partitions videos into semantically meaningful temporal chunks using a frozen video‑text backbone. By detecting changes in the embedding sequence of short video windows, STITCH produces reusable temporal abstractions that can be applied to multiple tasks such as event boundary detection, language‑based moment retrieval, and frame selection for vision‑language models. Experiments show that STITCH performs competitively with specialized methods while requiring no task‑specific training, especially when processing is limited to a few frames or tokens.