ITEP-84336: Fix incorrect camera pose for models generated with VGGT

manager/src/django/mesh_generator.py

@@ -26,6 +26,7 @@
 from scene_common.mesh_util import mergeMesh
 from scene_common.options import QUATERNION
 from scene_common import log
+from manager.serializers import CamSerializer
 
 ALLOWED_VIDEO_MIME_TYPES = {
     "video/mp4",
@@ -157,6 +158,7 @@ def startReconstructMesh(
     self,
     images: Dict[str, Dict],
     camera_order: List[str],
+    camera_location_order: List,
     mesh_type: str = "mesh",
     uploaded_map=None,
   ):
@@ -178,6 +180,10 @@ def startReconstructMesh(
         "mesh_type": mesh_type,
     }
 
+    camera_loc_by_id = {
+      cam_id: cam_loc
+      for cam_id, cam_loc in zip(camera_order, camera_location_order)
+    }
     log.info(f"Sending {len(images)} images to mapping service for reconstruction")
 
     files = []
@@ -201,6 +207,18 @@ def startReconstructMesh(
                 )
             )
             files.append(("camera_ids", (None, camera_id)))
+            cam_loc = camera_loc_by_id.get(camera_id)
+            if cam_loc is not None:
+              cam_loc = camera_loc_by_id.get(camera_id)
+              if cam_loc is not None:
+                cam_loc_clean = {
+                  "translation": list(cam_loc["translation"]),
+                  "rotation": list(cam_loc["rotation"]),
+                  "scale": list(cam_loc.get("scale", [1.0, 1.0, 1.0])),
+                }
+                files.append(("camera_locations", (None, json.dumps(cam_loc_clean))))
+            else:
+              log.warning(f"No camera location for {camera_id}")
           else:
             log.warning(
                 f"Camera {camera_id} in camera_order but not in images dict"
@@ -414,9 +432,30 @@ def startMeshGeneration(self, scene, mesh_type='mesh', uploaded_map=None):
       log.info(f"Collected {len(images)} images, calling mapping service")
       # Call mapping service to generate mesh
       # Pass camera IDs in order to ensure correct pose association
-      camera_order = [camera.sensor_id for camera in cameras]
+
+      camera_location_order = []
+      camera_order = []
+      serializer = CamSerializer()
+
+      for camera in cameras:
+        cam_id = camera.sensor_id
+        camera_order.append(cam_id)
+
+        t = serializer.get_translation(camera)
+        q = serializer.get_rotation(camera)
+        s = serializer.get_scale(camera) or [1.0, 1.0, 1.0]
+
+        if t is None or q is None:
+          raise ValueError(f"Missing pose for camera {cam_id}: t={t} q={q}")
+
+        camera_location_order.append({
+          "translation": list(t),
+          "rotation": list(q),
+          "scale": list(s),
+        })
+
       started = self.mapping_client.startReconstructMesh(
-        images, camera_order, mesh_type, uploaded_map_path
+        images, camera_order, camera_location_order, mesh_type, uploaded_map_path
       )
       rid = started.get("request_id")
       if not rid:

mapping/src/api_service_base.py

@@ -20,6 +20,7 @@
 from werkzeug.utils import secure_filename
 import uuid
 import threading
+import json
 
 from flask import Flask, request, jsonify
 from flask_cors import CORS
@@ -215,6 +216,7 @@ def reconstruct3D():
   image_files = request.files.getlist("images")
   video_file = request.files.get("video")
   camera_ids = request.form.getlist("camera_ids")
+  camera_locations = request.form.getlist("camera_locations", None)
 
   if (not image_files) and (video_file is None):
     set_status(request_id, state="failed", updated_at=time.time(), error="Provide images and/or video")
@@ -228,17 +230,29 @@ def reconstruct3D():
   images = None
   if image_files:
     images = []
-    pairs = zip(image_files, camera_ids) if camera_ids else [(f, None) for f in image_files]
-    for f, cam_id in pairs:
+
+    for idx, f in enumerate(image_files):
       if not f or not f.filename:
         continue
+
       raw = f.read()
       if not raw:
         continue
+
+      cam_id = camera_ids[idx] if idx < len(camera_ids) else None
+
+      cam_loc = None
+      if camera_locations and idx < len(camera_locations):
+        try:
+          cam_loc = json.loads(camera_locations[idx])
+        except Exception:
+          cam_loc = None
+
       images.append({
-        "filename": secure_filename(f.filename),
-        "camera_id": cam_id,
-        "data": base64.b64encode(raw).decode("utf-8"),
+          "filename": secure_filename(f.filename),
+          "camera_id": cam_id,
+          "camera_location": cam_loc,   # Only populated if provided
+          "data": base64.b64encode(raw).decode("utf-8"),
       })
 
     if not images: