"Wake Vision" is a large, high-quality dataset featuring over 6 million images, significantly exceeding the scale and diversity of current tinyML datasets (100x). This dataset includes images with annotations of whether each image contains a person. Additionally, it incorporates a comprehensive fine-grained benchmark to assess fairness and robustness, covering perceived gender, perceived age, subject distance, lighting conditions, and depictions. Annotations are published under a CC BY 4.0 license, and all images are sourced from the Open Images v7 dataset under a CC BY 2.0 license.
We provide a Colab to quickly interact with the dataset without downloading anything.
pip install -r requirements.txt
The dataset will automatically download if you use HuggingFace Datasets. Currently, you need to build the TFDS version of the dataset manually.
python3 benchmark.py -m"example_wake_vision_mobilenetv2.keras" -d=$DATASET_DIR -t={"tfds", "hf"}
python3 train.py -m="Wake_Vision_MNV2.keras" -d=$DATASET_DIR -t={"tfds", "hf"}