Refactor synap detector to more closely match other implementations

frigate/detectors/plugins/synap1680.py

@@ -1,35 +1,20 @@
 import logging
 
-import numpy as np
-from PIL import Image
-from pydantic import Field
 from typing_extensions import Literal
 
+import numpy as np
 from synap import Network
-from synap.types import Shape, Layout, DataType
-from synap.preprocessor import Preprocessor, InputData
+from synap.types import Shape, Layout
+from synap.preprocessor import Preprocessor
 from synap.postprocessor import Detector
 
 from frigate.detectors.detection_api import DetectionApi
-from frigate.detectors.detector_config import BaseDetectorConfig
-
-try:
-    from tflite_runtime.interpreter import Interpreter, load_delegate
-except ModuleNotFoundError:
-    from tensorflow.lite.python.interpreter import Interpreter, load_delegate
-
+from frigate.detectors.detector_config import BaseDetectorConfig, ModelTypeEnum
 
 logger = logging.getLogger(__name__)
 
 DETECTOR_KEY = "synap1680"
 
-# normalize the final bbox boundry
-def normalize_bbox(x, y, w, h, preproc_w, preproc_h) -> tuple[float, float, float, float]:
-    x_norm = x / preproc_w
-    y_norm = y / preproc_h
-    w_norm = (x + w) / preproc_w
-    h_norm = (y + h) / preproc_h
-    return x_norm, y_norm, w_norm, h_norm
 
 class SynapDetectorConfig(BaseDetectorConfig):
     type: Literal[DETECTOR_KEY]
@@ -41,63 +26,52 @@ class SynapDetector(DetectionApi):
 
         try:
             synap_network = Network(detector_config.model.path)
-            logger.info("[INIT] NPU found")
+            logger.info(f"Synap NPU loaded model: {detector_config.model.path}")
         except Exception as e:
-            logger.error(f"[INIT] Failed to init NPU: {e}")
+            logger.error(f"Failed to init Synap NPU: {e}")
             raise
         
+        self.width = detector_config.model.width
+        self.height = detector_config.model.height
+        self.model_type = detector_config.model.model_type
         self.network = synap_network
         self.network_input_details = self.network.inputs[0]
-        self.model_shape = self.network_input_details.shape
-        self.model_dtype = self.network_input_details.data_type
 
-    def detect_raw(self, tensor_input):
-        detections = np.zeros((20, 6), np.float32)
-        batch_size, height, width, channels = tensor_input.shape
-        target_height, target_width = self.model_shape[1], self.model_shape[2]
-        resized_tensors = []
-        for i in range(batch_size):
-            image = Image.fromarray(tensor_input[i])
+        # Create Inference Engine
+        self.preprocessor = Preprocessor()
+        self.detector = Detector(score_threshold=0.4, iou_threshold=0.4)
 
-            resized_image = image.resize((target_width, target_height))
-
-            resized_tensor = np.array(resized_image)
-            resized_tensors.append(resized_tensor)
-            
-            tensor_input = np.stack(resized_tensors)
-            if tuple(tensor_input.shape) != tuple(self.model_shape):
-                raise ValueError(f"diff shape {tensor_input.shape} {self.model_shape}")
-
-            preprocessor = Preprocessor()
-            detector = Detector()
-            assigned_rect = preprocessor.assign(self.network.inputs, tensor_input, Shape(tensor_input.shape), Layout.nhwc)
-            outputs = self.network.predict()
-            result = detector.process(outputs, assigned_rect)
+    def detect_raw(self, tensor_input: np.ndarray):
+        postprocess_data = self.preprocessor.assign(self.network.inputs, tensor_input, Shape(tensor_input.shape), Layout.nhwc)
+        output_tensor_obj = self.network.predict()
+        output = self.detector.process(output_tensor_obj, postprocess_data)
 
+        if self.model_type == ModelTypeEnum.yologeneric:
             detections = np.zeros((20, 6), np.float32)
 
-            for i, item in enumerate(result.items):
+            for i, item in enumerate(output.items):
                 if i == 20:
                     break
-                if float(item.confidence) < 0.4:
-                    continue
+
                 bb = item.bounding_box
-                raw_x, raw_y, raw_w, raw_h = normalize_bbox(
-                        bb.origin.x,
-                        bb.origin.y,
-                        bb.size.x,
-                        bb.size.y,
-                        assigned_rect.size.x,
-                        assigned_rect.size.y
-                )
+                
+                # Convert corner coordinates to normalized [0,1] range
+                x1 = bb.origin.x / self.width  # Top-left X
+                y1 = bb.origin.y / self.height  # Top-left Y
+                x2 = (bb.origin.x + bb.size.x) / self.width  # Bottom-right X
+                y2 = (bb.origin.y + bb.size.y) / self.height  # Bottom-right Y
+                                
                 detections[i] = [
                     item.class_index,
                     float(item.confidence),
-                    raw_y,
-                    raw_x,
-                    raw_h,
-                    raw_w,
+                    y1,
+                    x1,
+                    y2,
+                    x2,
                 ]
                 
             return detections
+        else:
+            print(f"Unsupported model type: {self.model_type}")
+            return np.zeros((20, 6), np.float32)