Add face detection feature on video input (#6)

Author: pierre-tenedero

README.md

@@ -19,9 +19,18 @@ mise run ws-e2e-chrome
 
 ## Run ws agent in browser
 
+### HAR model setup
+
 Download the onnx from https://modelnova.ai/models/details/human-activity-recognition ,
 and save it as `services/ws-server/static/models/human_activity_recognition.onnx`
 
+### Face detection setup
+
+Download the onnx from https://huggingface.co/amd/retinaface and save it in
+`services/ws-server/static/models/` and rename the file to `video_cv.onnx`.
+
+### Build and run the agent
+
 ```bash
 mise run build-ws-wasm-agent
 mise run ws-server
@@ -35,6 +44,7 @@ which will normally be something like 192.168.1.x.
 Then on your phone, open Chrome and type in https://192.168.1.x:8433/
 
 Click "Load HAR model" and then "Start sensors".
+For webcam inference, click "Load video CV model" and then "Start video".
 
 ## Grant
 

services/ws-server/src/main.rs

@@ -177,6 +177,27 @@ impl StreamHandler<Result<ws::Message, ws::ProtocolError>> for WebSocketActor {
                             action,
                             details,
                         } => {
+                            if capability == "video_cv" && action == "inference" {
+                                let detected_class = details
+                                    .get("detected_class")
+                                    .and_then(|value| value.as_str())
+                                    .unwrap_or("unknown");
+                                let confidence = details
+                                    .get("confidence")
+                                    .and_then(|value| value.as_f64())
+                                    .unwrap_or_default();
+                                let processed_at = details
+                                    .get("processed_at")
+                                    .and_then(|value| value.as_str())
+                                    .unwrap_or("unknown");
+                                info!(
+                                    "Video inference received from {}: class={} confidence={:.4} processed_at={}",
+                                    self.current_agent_id(),
+                                    detected_class,
+                                    confidence,
+                                    processed_at
+                                );
+                            }
                             info!(
                                 "Client event from {}: capability={} action={} details={}",
                                 self.current_agent_id(),