Quickstart: CLI¶

This guide walks through MindSight's command-line interface with progressive examples, from basic usage to full-featured tracking runs.

1. Basic Usage¶

Webcam (live):

python MindSight.py --source 0

Video file:

python MindSight.py --source video.mp4

Single image:

python MindSight.py --source image.jpg

Press Q to quit webcam or video playback. For images, press any key to close the window.

2. Adding Object Detection Classes¶

Filter detections to specific classes:

python MindSight.py --source video.mp4 --classes person knife cup

Or blacklist classes you want to ignore:

python MindSight.py --source video.mp4 --blacklist chair

3. Enabling Phenomena¶

Enable a single phenomenon:

python MindSight.py --source video.mp4 --joint-attention

Enable all phenomena at once:

python MindSight.py --source video.mp4 --all-phenomena

Add temporal confirmation windows to reduce false positives:

python MindSight.py --source video.mp4 \
  --joint-attention --ja-window 10 --ja-window-thresh 0.7

4. Saving Output¶

Save annotated video:

python MindSight.py --source video.mp4 --save

Log per-frame events to CSV:

python MindSight.py --source video.mp4 --log events.csv

Post-run summary CSV:

python MindSight.py --source video.mp4 --summary results.csv

Generate gaze heatmaps:

python MindSight.py --source video.mp4 --heatmap

Anonymize faces in the output video:

python MindSight.py --source video.mp4 --save --anonymize

Generate charts:

python MindSight.py --source video.mp4 --charts

5. Choosing a Gaze Backend¶

The default backend is MGaze. To use an alternative:

L2CS:

python MindSight.py --source video.mp4 --l2cs-model weights.pkl

UniGaze:

python MindSight.py --source video.mp4 --unigaze-model unigaze_b16_joint

Gazelle:

python MindSight.py --source video.mp4 --gazelle-model ckpt.pt

6. Using Visual Prompts¶

Supply a visual prompt file and a YOLOE model for open-vocabulary detection:

python MindSight.py --source video.mp4 \
  --vp-file prompt.vp.json --vp-model yoloe-26l-seg.pt

7. Tuning Gaze Parameters¶

Ray length (multiplier for the rendered gaze ray):

python MindSight.py --source video.mp4 --ray-length 1.5

Adaptive snap (snap the ray endpoint to the nearest object):

python MindSight.py --source video.mp4 --adaptive-ray snap --snap-dist 200

Gaze lock (hold gaze target for a dwell duration):

python MindSight.py --source video.mp4 --gaze-lock --dwell-frames 20

Gaze cone (widen the gaze hit-test angle, in degrees):

python MindSight.py --source video.mp4 --gaze-cone 5.0

8. Pipeline Configs¶

Load a full pipeline configuration from a YAML file:

python MindSight.py --pipeline my_pipeline.yaml

Pipeline configs let you define sources, detection settings, phenomena, gaze parameters, and outputs in a single file. See Project Mode for details.

9. Full Example¶

A complete command combining several features:

python MindSight.py --source video.mp4 \
  --classes person knife cup \
  --joint-attention --ja-window 10 \
  --mutual-gaze --social-ref \
  --adaptive-ray snap --snap-dist 200 \
  --save --summary results.csv --heatmap