Detection Results is the read-only feed of everything your AI models have detected. You’ll find it under AI & Detections → Detection Results in the sidebar. It shows detections as they happen (pushed live over a WebSocket) plus recent history from the database.

[!NOTE] This tab is shown to users with the AI Models permission (byom.manage). Anyone signed in can view results; the Delete and Delete Old actions are superuser-only.

What you can do here

• Watch detections stream in live — new rows appear the instant the backend produces them, with no polling.
• Filter by model, by task, and by result limit.
• Click a camera in the right-hand sidebar to focus on just that camera’s detections.
• Open a Camera Details dialog for per-camera statistics (total detections, models used, task types, average latency, recent detections).
• Refresh to reload history, and (superuser) delete individual results or bulk-delete old ones.

Read the results table

Each row shows one detection:

Column	Shows
ID	Result id (#…); live-pushed rows use a temporary negative id until you Refresh
Model	Model name and the camera it came from (Cam …)
Task	The task that produced it (e.g. person_detection)
Label	Detected class, or a caption for scene-description results
Confidence	Percentage, color-coded (green ≥ 80%, yellow ≥ 60%, orange below)
BBox	Bounding box: X, Y and width × height
Count	Object count (e.g. for person counting)
Latency	Inference latency in ms, color-coded (green < 200, yellow < 500, orange above)
Timestamp	Local date/time of the detection
Actions	Delete (superuser only)

[!NOTE] The table header counts the currently shown rows, e.g. Detection Results (100). Empty state reads “No detection results found. Run inference to generate results.” — start a model on AI Models (BYOM) to populate it.

Filter the feed

The filter bar sits above the table:

1. Filter by Model — pick a specific model or All Models.
2. Filter by Task — pick a task (options come from the tasks present in the current results) or All Tasks.
3. Limit — cap the number of rows: 50, 100, 200, or 500 (default 100).
4. Clear Filters resets model and task and returns the limit to 100.

Filters apply to both the loaded history and the live-pushed rows.

Focus on one camera

The right sidebar lists every camera with a badge showing how many results it has.

1. Click All Cameras to see everything, or click a camera to filter to it.
2. Clicking a camera also opens the Camera Details dialog, which shows:
   • Camera info (name, id, IP:port, status, manufacturer, model, location, active).
   • Detection statistics — Total Detections, Models Used, Task Types.
   • Models & Tasks running on that camera, each with its detection count and average latency.
   • Recent Detections — the last few for that camera.
3. Click Close (or outside the dialog) to dismiss it.

Refresh and clean up

• Refresh (top-right) reloads history from the server. Events that arrive mid-refresh are buffered so live rows aren’t lost.
• Delete (per row, superuser only) removes a single result after confirmation.
• Delete Old (7d+) (superuser only) bulk-deletes results older than 7 days after confirmation.

📸 Screenshots to capture

The existing infeencing results.jpeg is reused for the overview. Optionally add:

• detection-results-filters.png — the filter bar with a model and task selected.
• detection-results-camera-dialog.png — the Camera Details dialog open, showing the statistics tiles and Models & Tasks list.