added blur filtering

app.py

@@ -26,6 +26,8 @@ import numpy as np
 from PIL import Image
 import boto3
 
+from utils.blur_filter import filter_frames
+
 import os
 os.environ['KMP_DUPLICATE_LIB_OK'] = 'True'
 os.environ['OMP_NUM_THREADS'] = '4'
@@ -217,29 +219,20 @@ def process_video(vid_path, dsdg_thresh):
     cap = cv.VideoCapture(vid_path)
     input_width = int(cap.get(cv.CAP_PROP_FRAME_WIDTH))
     input_height = int(cap.get(cv.CAP_PROP_FRAME_HEIGHT))
-    fourcc = cv.VideoWriter_fourcc(*'mp4v')
-    output_vid_path = 'output_dsdg.mp4'
     
-    frame_counter = 0
-    all_frames = []
+    most_focused = filter_frames(cap)
+
     inference_images = []
     inference_bboxes = []
     inference_depths = []
-    while cap.isOpened():
-        ret, frame = cap.read()
-        if not ret:
-            break
-        # Process only every 5th frame
-        if frame_counter % 5 == 0:
-            # Run inference on the current frame
-            img, bbox, depth_img = analyze_face(frame)
-            if bbox and (depth_img is not None):
-                inference_images.append(img)
-                inference_bboxes.append(bbox)
-                inference_depths.append(depth_img)
-            all_frames.append(img)
-        frame_counter += 1
-    cap.release()
+    for frame in most_focused:
+        # Run inference on the current frame
+        img, bbox, depth_img = analyze_face(frame)
+        if bbox and (depth_img is not None):
+            inference_images.append(img)
+            inference_bboxes.append(bbox)
+            inference_depths.append(depth_img)
+    
     if not inference_images:
         return vid_path, {'Not supported right now': 0}, -1, vid_path, 'Faces too small or not found', -1
 
@@ -255,9 +248,11 @@ def process_video(vid_path, dsdg_thresh):
         text = f'{cls_dsdg} {w}*{h}'
         cv.rectangle(img, (x, y), (x2, y2), color_dsdg, 2)
         cv.putText(img, text, (x, y2 + 30), cv.FONT_HERSHEY_COMPLEX, 1, color_dsdg)
-
+    
+    fourcc = cv.VideoWriter_fourcc(*'mp4v')
+    output_vid_path = 'output_dsdg.mp4'
     out_dsdg = cv.VideoWriter(output_vid_path, fourcc, 6.0, (input_width, input_height))
-    for img in all_frames:
+    for img in most_focused:
         # Write the DSDG frame to the output video
         out_dsdg.write(img)
     out_dsdg.release()

utils/blur_filter.py

@@ -0,0 +1,53 @@
+from collections import deque
+import cv2 as cv
+import numpy as np
+
+
+def estimate_focus_measure(frame):
+    gray = cv.cvtColor(frame, cv.COLOR_BGR2GRAY)
+    laplacian = cv.Laplacian(gray, cv.CV_64F)
+    return laplacian.var()
+
+
+def update_threshold(blur_history):
+    return np.percentile(blur_history, 50)
+
+
+def select_most_focused(frame_history, threshold, select_n=5):
+    sorted_frames = sorted(frame_history, key=lambda x: x[1], reverse=True)
+    # Select least blurry frames from the last incomplete second
+    selected = [frame for frame, fm in sorted_frames[:select_n] if fm > threshold]
+    return selected
+
+
+def filter_frames(cap):
+    fps = int(cap.get(cv.CAP_PROP_FPS))
+    frame_history = deque(maxlen=fps)
+    blur_history = []
+    most_focused_frames = []
+
+    if cap.isOpened():
+        ret, frame = cap.read()
+        counter = 1
+        second = 0
+
+        while ret:
+            fm = estimate_focus_measure(frame)
+            frame_history.append([frame, fm])
+            blur_history.append(fm)
+
+            if counter >= fps:
+                second += 1
+                threshold = update_threshold(blur_history)
+                if counter % fps == 0:
+                    most_focused_frames += select_most_focused(frame_history, threshold)
+                    frame_history.clear()
+            ret, frame = cap.read()
+            counter += 1
+
+        if frame_history:
+            threshold = update_threshold(blur_history)
+            most_focused_frames += select_most_focused(frame_history, threshold)
+
+    cap.release()
+    return most_focused_frames