OnlineTAS: An Online Baseline for Temporal Action Segmentation

Authors: Qing Zhong, Guodong Ding, Angela Yao

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental On three common segmentation benchmarks, our approach achieves state-of-the-art performance.
Researcher Affiliation Academia Qing Zhong1,2 Guodong Ding2 Angela Yao2 1University of Adelaide 2National University of Singapore qing.zhong@adelaide.edu.au {dinggd,ayao}@comp.nus.edu.sg
Pseudocode Yes Algorithm 1 Adaptive Memory Update. ... Algorithm 2 Post-processing for Online TAS.
Open Source Code No The code will be made available once accepted.
Open Datasets Yes We evaluate our model on three common TAS datasets. Breakfast [18] comprises in total 1,712 videos... 50Salads [38] has 50 videos... GTEA [12] contains 28 videos...
Dataset Splits Yes We evaluate our model on three common TAS datasets. ... We train the model end-to-end with a learning rate of 5e 4 of total 50 epochs. Detailed hyperparameter settings can be found in Appendix.
Hardware Specification Yes We evaluate the runtime performance of our approach using an Nvidia A40 GPU... A single NVIDIA RTX 3090 GPU is sufficient for training.
Software Dependencies No The paper mentions several architectural components (e.g., GRU, Transformer, Swin, TCN) and implies the use of common machine learning libraries, but it does not specify software dependencies with version numbers (e.g., 'PyTorch 1.9' or 'CUDA 11.1').
Experiment Setup Yes We train the model end-to-end with a learning rate of 5e 4 of total 50 epochs. Detailed hyperparameter settings can be found in Appendix. ... Table 11: Hyper-parameter settings for GTEA, 50Salads, and Breakfast datasets.