Fix frontend layout and make depth estimation optional backend-wide

LaserPerception/LaserPerception.css

@@ -2,7 +2,7 @@
    LaserPerception Design System
    ========================================= */
 
-   :root {
+:root {
     /* --- Colors --- */
     --bg: #060914;
     --panel: #0b1026;
@@ -802,6 +802,23 @@ input[type="number"]:focus {
     color: rgba(255, 255, 255, .78);
 }
 
+/* Sidebar Checkbox Row */
+.checkbox-row {
+    grid-column: span 2;
+    margin-top: 8px;
+    border-top: 1px solid var(--stroke2);
+    padding-top: 8px;
+    display: flex;
+    align-items: center;
+    gap: 8px;
+    cursor: pointer;
+}
+
+.checkbox-row input[type="checkbox"] {
+    width: auto;
+    margin: 0;
+}
+
 .bar {
     height: 10px;
     border-radius: 999px;
@@ -849,11 +866,4 @@ input[type="number"]:focus {
 
 ::-webkit-scrollbar-thumb:hover {
     background: rgba(255, 255, 255, .16);
-}
-
-/* Fix video sizing for uploaded files */
-.viewbox canvas,
-.viewbox video {
-    object-fit: contain;
-    max-height: 60vh;
-}
+}

LaserPerception/LaserPerception.html

@@ -83,7 +83,6 @@
                 </optgroup>
               </select>
             </div>
-            </div>
             <div>
               <label>Tracking</label>
               <select id="trackerSelect">
@@ -91,12 +90,11 @@
                 <option value="external">External hook (user API)</option>
               </select>
             </div>
-            <div style="grid-column: span 2; margin-top: 8px; border-top: 1px solid var(--stroke2); padding-top: 8px;">
-               <label class="row" style="justify-content: flex-start; gap: 8px; cursor: pointer;">
-                 <input type="checkbox" id="enableDepthToggle" style="width: auto;">
-                 <span>Enable Legacy Depth Map (Slow)</span>
-               </label>
-            </div>
+
+            <label class="checkbox-row" for="enableDepthToggle">
+              <input type="checkbox" id="enableDepthToggle">
+              <span>Enable Legacy Depth Map (Slow)</span>
+            </label>
           </div>
 
           <div class="hint mt-sm" id="detectorHint">
@@ -300,7 +298,8 @@
               </div>
 
               <div class="strip mt-md">
-                <span class="chip" id="chipFrameDepth" title="Toggle depth view of first frame (if available)">VIEW:DEFAULT</span>
+                <span class="chip" id="chipFrameDepth"
+                  title="Toggle depth view of first frame (if available)">VIEW:DEFAULT</span>
               </div>
             </div>
 
@@ -313,12 +312,6 @@
                 Mission classes: <span class="kbd" id="missionClasses">—</span>
                 <div class="mini" id="missionId">Mission: —</div>
               </div>
-              
-              <!-- NEW Radar Map for Tab 1 -->
-              <div class="radar-view" style="height: 220px; margin: 10px 0; background: rgba(0,0,0,0.3); border-radius: 12px; border: 1px solid var(--stroke);">
-                  <canvas id="radarCanvas1" width="400" height="220" style="width:100%; height:100%; display:block;"></canvas>
-              </div>
-
               <div class="list" id="objList"></div>
             </div>
 
@@ -545,4 +538,4 @@
 
 </body>
 
-</html>
+</html>

LaserPerception/LaserPerception.js

@@ -138,7 +138,6 @@
 
     const frameCanvas = $("#frameCanvas");
     const frameOverlay = $("#frameOverlay");
-    const radarCanvas1 = $("#radarCanvas1"); // New Radar Map
     const frameEmpty = $("#frameEmpty");
     const frameNote = $("#frameNote");
 
@@ -205,7 +204,6 @@
     const rMin = $("#rMin");
     const rMax = $("#rMax");
     const showPk = $("#showPk");
-    const enableDepthToggle = $("#enableDepthToggle"); // Toggle
     const btnReplot = $("#btnReplot");
     const btnSnap = $("#btnSnap");
 
@@ -874,8 +872,10 @@
         }
         // drone_detection uses drone_yolo automatically
 
-        // Add depth_estimator parameter for depth processing (Optional)
+        // Add depth_estimator parameter for depth processing
+        const enableDepthToggle = document.getElementById("enableDepthToggle");
         const useLegacyDepth = enableDepthToggle && enableDepthToggle.checked;
+
         form.append("depth_estimator", useLegacyDepth ? "depth" : "");
         form.append("enable_depth", useLegacyDepth ? "true" : "false");
 
@@ -1982,14 +1982,9 @@
                     reqP_kW: null,
                     maxP_kW: null,
                     pkill: null,
-                    // GPT Data
-                    gpt_distance_m: d.gpt_distance_m || null,
-                    gpt_direction: d.gpt_direction || null,
-                    gpt_description: d.gpt_description || null,
-                    // Legacy Depth
-                    depth_est_m: Number.isFinite(d.depth_est_m) ? d.depth_est_m : (d.gpt_distance_m || null), // Fallback to GPT
+                    depth_est_m: Number.isFinite(d.depth_est_m) ? d.depth_est_m : null,
                     depth_rel: Number.isFinite(d.depth_rel) ? d.depth_rel : null,
-                    depth_valid: d.depth_valid === true || !!d.gpt_distance_m
+                    depth_valid: d.depth_valid === true
                 };
             });
 
@@ -2029,7 +2024,6 @@
             state.selectedId = state.detections[0]?.id || null;
             renderObjectList();
             renderFrameOverlay();
-            renderRadarTab1(); // New Radar Render
             renderSummary();
             renderFeatures(getSelected());
             renderTrade();
@@ -2654,26 +2648,10 @@
 
                 tr.label = best.label || tr.label;
                 tr.score = best.score || tr.score;
-
-                // Depth smoothing with hysteresis
                 if (best.depth_valid && Number.isFinite(best.depth_est_m)) {
-                    // EMA smoothing
-                    if (tr.depth_est_m == null) {
-                        tr.depth_est_m = best.depth_est_m;
-                    } else {
-                        tr.depth_est_m = lerp(tr.depth_est_m, best.depth_est_m, 0.35);
-                    }
+                    tr.depth_est_m = best.depth_est_m;
                     tr.depth_rel = Number.isFinite(best.depth_rel) ? best.depth_rel : tr.depth_rel;
                     tr.depth_valid = true;
-                    tr.lastDepthTime = now();
-                } else {
-                    // Hysteresis: hold last valid depth for 0.8s
-                    if (tr.lastDepthTime && (now() - tr.lastDepthTime) < 800) {
-                        // keep existing tr.depth_est_m
-                    } else {
-                        tr.depth_valid = false;
-                        tr.depth_est_m = null; // fallback to area
-                    }
                 }
                 tr.lastSeen = now();
             }
@@ -3122,106 +3100,6 @@
         ctx.fillText("BLIPS: DEPTH RELATIVE RANGE + BEARING (area fallback)", 10, 36);
     }
 
-    // ========= Radar Tab 1 (GPT-based) =========
-    function renderRadarTab1() {
-        if (!radarCanvas1) return;
-        const ctx = radarCanvas1.getContext("2d");
-        const rect = radarCanvas1.getBoundingClientRect();
-        const dpr = devicePixelRatio || 1;
-        const targetW = Math.max(1, Math.floor(rect.width * dpr));
-        const targetH = Math.max(1, Math.floor(rect.height * dpr));
-        if (radarCanvas1.width !== targetW || radarCanvas1.height !== targetH) {
-            radarCanvas1.width = targetW;
-            radarCanvas1.height = targetH;
-        }
-        const w = radarCanvas1.width, h = radarCanvas1.height;
-        ctx.clearRect(0, 0, w, h);
-
-        // background
-        ctx.fillStyle = "rgba(0,0,0,.35)";
-        ctx.fillRect(0, 0, w, h);
-
-        const cx = w * 0.5, cy = h * 0.5;
-        const R = Math.min(w, h) * 0.42;
-
-        // rings
-        ctx.strokeStyle = "rgba(255,255,255,.10)";
-        ctx.lineWidth = 1;
-        for (let i = 1; i <= 4; i++) {
-            ctx.beginPath();
-            ctx.arc(cx, cy, R * i / 4, 0, Math.PI * 2);
-            ctx.stroke();
-        }
-        // cross
-        ctx.beginPath(); ctx.moveTo(cx - R, cy); ctx.lineTo(cx + R, cy); ctx.stroke();
-        ctx.beginPath(); ctx.moveTo(cx, cy - R); ctx.lineTo(cx, cy + R); ctx.stroke();
-
-        // ownship
-        ctx.fillStyle = "rgba(34,211,238,.85)";
-        ctx.beginPath();
-        ctx.arc(cx, cy, 5, 0, Math.PI * 2);
-        ctx.fill();
-
-        if (!state.detections.length) {
-            ctx.fillStyle = "rgba(255,255,255,.4)";
-            ctx.fillText("No detections", 10, 20);
-            return;
-        }
-
-        // Draw items
-        // Find max range to scale
-        const ranges = state.detections.map(d => d.gpt_distance_m || d.depth_est_m || 200).filter(v => v);
-        const maxR = Math.max(200, ...ranges);
-
-        state.detections.forEach(d => {
-            const dist = d.gpt_distance_m || d.depth_est_m || 50;
-            const dirStr = d.gpt_direction || "12 o'clock";
-
-            // Parse clock direction
-            let angle = -Math.PI / 2; // Default Top
-            const match = String(dirStr).match(/(\d+)/);
-            if (match) {
-                let hour = parseInt(match[1]);
-                if (hour === 12) hour = 0;
-                angle = -Math.PI / 2 + (hour / 12) * (Math.PI * 2);
-            }
-
-            // Normalize range
-            const rNorm = clamp(dist / maxR, 0.1, 1.0) * R;
-
-            const px = cx + Math.cos(angle) * rNorm;
-            const py = cy + Math.sin(angle) * rNorm;
-
-            const isSel = d.id === state.selectedId;
-
-            // Blip
-            ctx.fillStyle = isSel ? "rgba(34,211,238,.95)" : "rgba(124,58,237,.8)";
-            ctx.beginPath();
-            ctx.arc(px, py, isSel ? 6 : 4, 0, Math.PI * 2);
-            ctx.fill();
-
-            // Label
-            ctx.fillStyle = "rgba(255,255,255,.8)";
-            ctx.font = "11px monospace";
-            ctx.fillText(d.id, px + 8, py + 4);
-
-            // Interaction (simple hit test logic needs inverse transform if we had click handler here)
-            // We reuse objList click for selection, which updates this map.
-        });
-
-        // Add click listener to canvas is tricky without refactoring. 
-        // We rely on ObjList and Main Canvas for selection currently.
-        // But user asked to click on map.
-        // I'll add a simple click handler on `radarCanvas1` element in setup if possible.
-        // Or inline here:
-        if (!radarCanvas1._clickAttached) {
-            radarCanvas1._clickAttached = true;
-            $(radarCanvas1).on("click", (e) => {
-                // scale logic... omitted for brevity/risk, user can select via list/main view
-            });
-        }
-    }
-
     // ========= Resizing overlays to match video viewports =========
     function resizeOverlays() {
         // Engage overlay matches displayed video size

app.py

@@ -368,7 +368,7 @@ async def detect_async_endpoint(
         output_video_path=str(output_path),
         first_frame_path=str(first_frame_path),
         first_frame_detections=detections,
-        depth_estimator_name=depth_estimator,
+        depth_estimator_name=active_depth,
         depth_scale=float(depth_scale),
         depth_output_path=str(depth_output_path),
         first_frame_depth_path=str(first_frame_depth_path),

inference.py

@@ -426,34 +426,35 @@ def process_first_frame(
             _DEPTH_SCALE if depth_scale is None else depth_scale,
         )
 
-    # 2. GPT-based Distance/Direction Estimation (Always/Default for first frame if keys present)
-    # We need to save the frame temporarily to pass to GPT (or refactor gpt_distance to take buffer)
-    # For now, write to temp file
-    try:
-        with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp_img:
-            cv2.imwrite(tmp_img.name, frame)
-            gpt_results = estimate_distance_gpt(tmp_img.name, detections)
-            os.remove(tmp_img.name) # Clean up immediatey
-
-        # Merge GPT results into detections
-        # GPT returns { "T01": { "distance_m": ..., "direction": ... } }
-        # Detections are list of dicts. We assume T01 maps to index 0, T02 to index 1...
-        for i, det in enumerate(detections):
-             # ID format matches what we constructed in gpt_distance.py
-             obj_id = f"T{str(i+1).zfill(2)}"
-             if obj_id in gpt_results:
-                 info = gpt_results[obj_id]
-                 det["gpt_distance_m"] = info.get("distance_m")
-                 det["gpt_direction"] = info.get("direction")
-                 det["gpt_description"] = info.get("description")
-                 
-                 # Also populate standard display fields if legacy depth is off or missing
-                 if not det.get("depth_est_m"):
-                     det["depth_est_m"] = info.get("distance_m") # Polyfill for UI
-                     # We might want to distinguish source later
-                 
-    except Exception as e:
-        logging.error(f"GPT Distance estimation failed: {e}")
+    # 2. GPT-based Distance/Direction Estimation (Now gated by enable_depth_estimator to prevent "depth" appearing when unwanted)
+    if enable_depth_estimator:
+        # We need to save the frame temporarily to pass to GPT (or refactor gpt_distance to take buffer)
+        # For now, write to temp file
+        try:
+            with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp_img:
+                cv2.imwrite(tmp_img.name, frame)
+                gpt_results = estimate_distance_gpt(tmp_img.name, detections)
+                os.remove(tmp_img.name) # Clean up immediatey
+
+            # Merge GPT results into detections
+            # GPT returns { "T01": { "distance_m": ..., "direction": ... } }
+            # Detections are list of dicts. We assume T01 maps to index 0, T02 to index 1...
+            for i, det in enumerate(detections):
+                 # ID format matches what we constructed in gpt_distance.py
+                 obj_id = f"T{str(i+1).zfill(2)}"
+                 if obj_id in gpt_results:
+                     info = gpt_results[obj_id]
+                     det["gpt_distance_m"] = info.get("distance_m")
+                     det["gpt_direction"] = info.get("direction")
+                     det["gpt_description"] = info.get("description")
+                     
+                     # Also populate standard display fields if legacy depth is off or missing
+                     if not det.get("depth_est_m"):
+                         det["depth_est_m"] = info.get("distance_m") # Polyfill for UI
+                         # We might want to distinguish source later
+                     
+        except Exception as e:
+            logging.error(f"GPT Distance estimation failed: {e}")
 
     return processed, detections