Helios Data Visualizer

A tool for rendering sensor data from the Helios device, as provided in the CMI - Detect Behavior with Sensor Data competition featured on Kaggle. A discussion post for the tool can be found here.

The visualizer retrieves all sequences from the dataset that match the given filters. A video file (.mp4) is then produced for the sequence at the index specified by --sequence_index (default is 0, the first match).

Example Video Frame:

Watch an example video (YouTube)

Installation:

1. Create a python virtual environment (recommended)

2. Install dependencies: pip install -r requirements.txt

3. Install FFmpeg (required for video generation):

   • Windows: Download from https://ffmpeg.org/download.html and add to PATH
   • macOS: brew install ffmpeg
   • Linux: sudo apt install ffmpeg

Example Usage

Render the first sequence found in the dataset:

python helios_render.py --csv cmi-detect-behavior-with-sensor-data/train.csv

Render all "Wave hello" gestures for subject SUBJ_032761:

python helios_render.py --csv cmi-detect-behavior-with-sensor-data/train.csv --subject SUBJ_032761 --gesture "Wave hello"

Render the 4th "Neck - pinch skin" gesture for subject SUBJ_032761:

python helios_render.py --csv cmi-detect-behavior-with-sensor-data/train.csv --subject SUBJ_032761 --gesture "Neck - pinch skin" --sequence_index 3

Arguments

Argument	Required	Usage
--csv	✅	Path to the CSV data file
--subject		Subject ID to filter (e.g., SUBJ_059520)
--gesture		Gesture description to filter
--behavior		Behavior description to filter
--phase		Phase to filter
--orientation		Orientation to filter
--output_dir		Output directory for videos
--framerate		Frame rate for output video
--sequence_index		Index of the sequence to render (if multiple sequences match)
--no-cache		Disable parquet caching