TTA-Vid: Generalized Test-Time Adaptation for Video Reasoning

2026-04-01

AI Summary

This work addresses the challenge of generalizing video reasoning models to new domains, which typically rely on large-scale annotated data and multi-stage training. The study introduces, for the first time, test-time reinforcement learning to video-language tasks, proposing a label-free adaptive approach that operates effectively under single-sample or single-batch conditions. By progressively reasoning over subsets of video frames, the method integrates a batch-aware frequency-based reward mechanism with a multi-armed bandit strategy for keyframe selection. This enables robust cross-dataset generalization without access to ground-truth labels during testing. Empirical results demonstrate consistent performance gains over state-of-the-art models that depend on extensive training data, significantly improving both adaptation efficiency and accuracy at test time across diverse video reasoning benchmarks.

Abstract

Recent video reasoning models have shown strong results on temporal and multimodal understanding, yet they depend on large-scale supervised data and multi-stage training pipelines, making them costly to train and difficult to adapt to new domains. In this work, we leverage the paradigm of Test-Time Reinforcement Learning on video-language data to allow for adapting a pretrained model to incoming video samples at test-time without explicit labels. The proposed test-time adaptation for video approach (TTA-Vid) combines two components that work simultaneously: (1) a test-time adaptation that performs step-by-step reasoning at inference time on multiple frame subsets. We then use a batch-aware frequency-based reward computed across different frame subsets as pseudo ground truth to update the model. It shows that the resulting model trained on a single batch or even a single sample from a dataset, is able to generalize at test-time to the whole dataset and even across datasets. Because the adaptation occurs entirely at test time, our method requires no ground-truth annotations or dedicated training splits. Additionally, we propose a multi-armed bandit strategy for adaptive frame selection that learns to prioritize informative frames, guided by the same reward formulation. Our evaluation shows that TTA-Vid yields consistent improvements across various video reasoning tasks and is able to outperform current state-of-the-art methods trained on large-scale data. This highlights the potential of test-time reinforcement learning for temporal multimodal understanding.

Problem

• video reasoning
• test-time adaptation
• temporal understanding
• multimodal learning
• domain adaptation

Innovation

• Test-Time Adaptation
• Reinforcement Learning
• Video Reasoning
• Frame Selection
• Multimodal Understanding