lei-qi-233/MicroG-HAR-train-ready

MicroG-HAR-train-ready dataset.

This dataset is converted from the MicroG-4M dataset.

For more information please:

• —Refer to our [paper](https://arxiv.org/abs/2506.02845)

• —Visit our [GitHub](https://github.com/LEI-QI-233/HAR-in-Space)

Datasets formatted in this way can be used directly as input data directories for the PySlowFast_for_HAR framework without additional preprocessing, enabling seamless model training and evaluation.

This dataset follows the organizational format of the AVA dataset. The only difference is that the original CSV header middle_frame_timestamp (the timestamp in seconds from the start of the video) has been replaced with key_frame_stamp, which stores the index of the key frame. The key frame index is defined as the index of the middle frame among all annotated bounding‐box frames for the same person_id within any continuous three‐second (90‐frame) segment.

Folders and Files in Repository

ava Folder:

contains all data files for fine-tuning on PySlowFast_for_HAR.

NOTE: Please unzip the frames.zip file before using.

ava_with_head Folder:

contains same ava files with header.

configs Folder:

contains configuration files for fine-tuning on PySlowFast_for_HAR.

How to use it

Please see DATASET.md.