test

app.py

@@ -8,11 +8,13 @@ from moviepy.video.io.ImageSequenceClip import ImageSequenceClip
 from configuration import Config
 from model import load_classifier, load_detector
 from inference import format_frame, detect_object, classify_action, draw_boxes
+
 config = Config()
 print(f'TensorFlow {tf.__version__}')
 
 print(f'Load classifier from {config.classifier_path}')
 classifier = load_classifier(config)
+classifier.trainable = False
 classifier.summary()
 
 print('Load detector.')
@@ -28,9 +30,9 @@ def fn(video: gr.Video):
     actions = []
     detections = ([], [])
     for i, frame in enumerate(clip.iter_frames()):
-      if i % config.classify_action_frame_steps == 0:
+      if i % config.classify_action_frame_step == 0:
         frames.append(format_frame(frame, config))
-      if i % config.detect_object_frame_steps == 0:
+      if i % config.detect_object_frame_step == 0:
         print(f'Detect object: Frame {i}')
         detections = detect_object(detector, frame)
       if len(frames) == config.classify_action_num_frames:
@@ -40,10 +42,10 @@ def fn(video: gr.Video):
       frame = draw_boxes(frame, detections, actions)
       processed_frames.append(frame)
       if i % config.yield_frame_steps == 0:
+        quality = 9
+        image_array = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
+        _, image_encoded = cv2.imencode('.jpg', image_array, [int(cv2.IMWRITE_JPEG_QUALITY), quality])
         with tempfile.NamedTemporaryFile(suffix='.jpeg') as f:
-          quality = 9
-          image_array = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
-          _, image_encoded = cv2.imencode('.jpg', image_array, [int(cv2.IMWRITE_JPEG_QUALITY), quality])
           f.write(image_encoded)
           yield f.name, None
     processed_clip = ImageSequenceClip(processed_frames, clip.fps)

configuration.py

@@ -1,5 +1,5 @@
 class Config:
-  num_frames = 8
+  num_frames = 4
   frame_step = 15
   resolution = 224
   frame_size = (resolution, resolution)
@@ -26,7 +26,7 @@ class Config:
   num_classes = len(id_to_name)
   input_shape = (1, num_frames, resolution, resolution, 3)
 
-  detect_object_frame_steps = 5
-  classify_action_frame_steps = 15
-  classify_action_num_frames = 8
+  detect_object_frame_step = 5
+  classify_action_frame_step = frame_step
+  classify_action_num_frames = num_frames
   yield_frame_steps = 5

inference.py

@@ -1,6 +1,6 @@
-from imgviz import instances2rgb
 import tensorflow as tf
 import numpy as np
+from imgviz import instances2rgb
 
 from configuration import Config
 
@@ -18,8 +18,7 @@ def detect_object(detector, frame):
   boxes = result.boxes.xyxy.numpy()
   detections = (
     [result.names[i].capitalize() for i in classes],
-    boxes,
-  )
+    boxes)
   return detections
 
 def classify_action(classifier, frames, id_to_name):
@@ -48,12 +47,10 @@ def draw_boxes(frame, detections, actions):
   line_width = 2
   captions = [
     f'{class_name}\n' + '\n'.join(actions if i == max_area_id else [])
-    for (i, class_name) in enumerate(classes)
-  ]
+    for (i, class_name) in enumerate(classes)]
   bboxes = [
     [box[1], box[0], box[3], box[2]]
-    for box in boxes
-  ]
+    for box in boxes]
   frame = instances2rgb(
     frame,
     labels=labels,
@@ -61,8 +58,7 @@ def draw_boxes(frame, detections, actions):
     bboxes=bboxes,
     colormap=colormap,
     font_size=20,
-    line_width=line_width,
-  )
+    line_width=line_width)
   return frame
 
 def FrameProcessor(detector, classifier, config: Config):

model.py

@@ -48,7 +48,3 @@ def load_classifier(config: Config):
 
 def load_detector(config: Config):
   return YOLO(config.detector_path)
-
-def compile_classifier(model, config: Config):
-  optimizer = keras.optimizers.Adam(learning_rate=config.learning_rate)
-  model.compile(optimizer=optimizer, loss='sparse_categorical_crossentropy', metrics=['accuracy'])