made raw mask generation mode to predict the same class

supervisely_integration/serve/config.json

@@ -20,7 +20,7 @@
     ],
     "community_agent": false,
     "docker_image": "supervisely/segment-anything-hq:1.0.5",
-    "instance_version": "6.7.33",
+    "instance_version": "6.9.22",
     "entrypoint": "python -m uvicorn main:model.app --app-dir ./supervisely_integration/serve/src --host 0.0.0.0 --port 8000 --ws websockets",
     "port": 8000,
     "task_location": "application_sessions",
@@ -29,8 +29,8 @@
     },
     "access_restriction": [
         {
-          "instance": "community_free",
-          "message": "The best interactive models for AI-assisted segmentation are already deployed in the Supervisely Cloud and are available for auto-labeling to all platform users. The number of API calls and the ability to run this app on your own agent (GPU) are limited for Free accounts. To increase usage limits or run the app on your GPU, switch to the <a href=\"/billing\">Pro</a> plan or request the <a href=\"https://supervisely.com/contact-us/\" target=\"_blank\">Enterprise</a> Edition."
+            "instance": "community_free",
+            "message": "The best interactive models for AI-assisted segmentation are already deployed in the Supervisely Cloud and are available for auto-labeling to all platform users. The number of API calls and the ability to run this app on your own agent (GPU) are limited for Free accounts. To increase usage limits or run the app on your GPU, switch to the <a href=\"/billing\">Pro</a> plan or request the <a href=\"https://supervisely.com/contact-us/\" target=\"_blank\">Enterprise</a> Edition."
         }
-    ]    
-}
+    ]
+}

supervisely_integration/serve/src/main.py

@@ -108,7 +108,7 @@ def load_on_device(
         # build predictor
         self.predictor = SamPredictor(self.sam)
         # define class names
-        self.class_names = ["target_mask"]
+        self.class_names = ["object_mask"]
         # list for storing mask colors
         self.mask_colors = [[255, 0, 0]]
         # variable for storing image ids from previous inference iterations
@@ -180,18 +180,10 @@ def predict(self, image_path: str, settings: Dict[str, Any]) -> List[sly.nn.Pred
                 output_mode=settings["output_mode"],
             )
             masks = mask_generator.generate(input_image)
-            for i, mask in enumerate(masks):
-                class_name = "object_" + str(i)
-                # add new class to model meta if necessary
-                if not self._model_meta.get_obj_class(class_name):
-                    color = generate_rgb(self.mask_colors)
-                    self.mask_colors.append(color)
-                    self.class_names.append(class_name)
-                    new_class = sly.ObjClass(class_name, sly.Bitmap, color)
-                    self._model_meta = self._model_meta.add_obj_class(new_class)
+            for mask in masks:
                 # get predicted mask
                 mask = mask["segmentation"]
-                predictions.append(sly.nn.PredictionMask(class_name=class_name, mask=mask))
+                predictions.append(sly.nn.PredictionMask(class_name="object_mask", mask=mask))
         elif settings["mode"] == "bbox":
             # get bbox coordinates
             if "rectangle" not in settings: