Collect a reusable dataset
Save original frames and separate YOLO labels. Keep annotation previews out of the source images so the dataset can be reviewed and relabeled.
Select a session and sample cap
collection.yaml defines sessions tied to enabled cameras. Choose interval, random, or on-detection sampling; add class/confidence filters and a UTC schedule when needed. max_frames counts saved images across restarts, not objects.
Access files directly on the host
With the default output directory, open collected/<session_id> in the checkout. Inside the container, that folder is mounted under /data/collected.
collected/entrance_dataset/
data.yaml
images/<sample_id>.jpg
labels/<sample_id>.txt
metadata/ # optional JSON sidecars
.collector.sqlite3 # saved counts and recovery state
.collector.lock
.pending/
.quarantine/Export or continue collecting
Stop collection before exporting images, labels, and data.yaml. Keep the full session folder, including its journal, to resume safely. At the sample cap the collector remains completed and healthy; increase the cap and recreate/restart it to continue.