AViD Dataset: Anonymized Videos from Diverse Countries

Authors: AJ Piergiovanni, Michael Ryoo

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally illustrate that models trained with such biased datasets do not transfer perfectly to action videos from the other countries, and show that AVi D addresses such problem. We also confirm that the new AVi D dataset could serve as a good dataset for pretraining the models, performing comparably or better than prior datasets. We conducted a series of experiments with the new AVi D dataset.
Researcher Affiliation Academia AJ Piergiovanni Indiana University ajpiergi@indiana.edu Michael S. Ryoo Stony Brook University mryoo@cs.stonybrook.edu
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code Yes The dataset is available https://github.com/piergiaj/AVi D. The models... were trained for 256 epochs... (code provided in supplementary materials).
Open Datasets Yes We introduce a new public video dataset for action recognition: Anonymized Videos from Diverse countries (AVi D). The dataset is available https://github.com/piergiaj/AVi D.
Dataset Splits No The paper states: 'We split the dataset into train/test sets by taking 10% of each class as the test videos.' It does not explicitly mention a 'validation' split or provide details for one.
Hardware Specification Yes We implemented the models in Py Torch and trained them using four Titan V GPUs.
Software Dependencies No The paper mentions 'Py Torch' but does not specify a version number or other software dependencies with versions.
Experiment Setup Yes The models... were trained for 256 epochs. The learning rate followed a cosine decay schedule with a max of 0.1 and a linear warm-up for the first 2k steps. Each GPU used a base batch size of 8 clips... The base clip size was 32 frames at 224 224 image resolution.