Reproduce “Calibrating Multimodal Test-Time Adaptation under Reliability and Availability Modality Shifts” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiangmeng Li, Jinglin Xu, Yi Li, Yumeng Liu, Jiahuan Zhou, Fanjiang Xu, Hui Xiong, Gang Hua
The paper presents AdaPGC for adapting audio-video classifiers to noisy or missing modalities at test time. It models Gaussian distributions over modality and fused features to calibrate predictions, flag unreliable inputs, and estimate missing fusion features. On Kinetics50-C, its average across 15 video corruptions is 66.7% versus 65.9%; at 90% Both-Missing, it scores 61.1% versus 59.5%. These are benchmark accuracy comparisons, not wins in every setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.