Action Sequence Augmentation for Action Anticipation

Authors: Yihui Qiu, Deepu Rajan

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments on the 50Salads, EGTEA Gaze+, and Epic-Kitchens-100 datasets demonstrate significant performance improvements over existing state-of-the-art methods.
Researcher Affiliation Academia Yihui Qiu & Deepu Rajan College of Computing and Data Science Nanyang Technological University, Singapore EMAIL
Pseudocode No The paper describes the Action Grammar Induction (AGI) algorithm and the Cross-Tree Earley Parser (CTEP) operations in text and bullet points, but does not provide structured pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements about releasing code or links to source code repositories for the methodology described.
Open Datasets Yes We perform experiments on three action anticipation benchmarks. 50Salads Stein & Mc Kenna (2013) consists of 901 action annotations, and 17 action classes. EGTEA Gaze+ Li et al. (2021) contains 10,325 action annotations, 19 verbs, 51 nouns and 106 action classes. EPIC-Kitchens-100 Damen et al. (2022) contains 3806 actions, with 97 verbs, and 300 nouns.
Dataset Splits Yes We report the average performance across the standard five splits. Methods are evaluated on EGTEA Gaze+ reporting the average performance across the three splits provided by the authors of the dataset. We evaluate our method on the validation dataset following previous work Guo et al. (2024).
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup No The paper discusses augmentation strategies and their impact on performance, but does not explicitly provide concrete hyperparameter values or detailed training configurations (e.g., learning rate, batch size, number of epochs, optimizer settings) for the models used in the experiments.