Merge pull request #366 from grycap/dev-srisco

OSCAR2 integration & mask-detector-workflow example

Author: srisco

examples/mask-detector-workflow/blurry-faces.sh

@@ -0,0 +1,16 @@
+#!/bin/bash
+
+VIDEO_NAME=`basename "$INPUT_FILE_PATH"`
+SUBFOLDER_NAME=`echo "$VIDEO_NAME" | cut -f 1 -d '.'`
+OUTPUT_SUBFOLDER="$TMP_OUTPUT_DIR/$SUBFOLDER_NAME"
+
+mkdir "$OUTPUT_SUBFOLDER"
+
+echo "SCRIPT: Analyzing file '$INPUT_FILE_PATH', saving the output images in '$OUTPUT_SUBFOLDER'"
+
+ffmpeg -i "$INPUT_FILE_PATH" -vf fps=12/60 "$OUTPUT_SUBFOLDER/img%d.jpg"
+
+for IMAGE in "$OUTPUT_SUBFOLDER"/*
+do 
+	python auto_blur_image.py -i "$IMAGE" -o "$IMAGE"
+done

examples/mask-detector-workflow/blurry-faces/Dockerfile

@@ -0,0 +1,15 @@
+FROM python:slim-buster
+
+RUN pip install --no-cache-dir opencv-python numpy tensorflow && \
+    rm -rf /root/.cache/pip/* && \
+    rm -rf /tmp/*
+
+RUN apt update && \
+    apt install -y --no-install-recommends ffmpeg && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
+COPY . /opt/blurry-faces
+
+WORKDIR /opt/blurry-faces/src
+

examples/mask-detector-workflow/blurry-faces/src/DetectorAPI.py

@@ -0,0 +1,64 @@
+import tensorflow as tf
+import numpy as np
+import time
+
+class DetectorAPI:
+    def __init__(self, path_to_ckpt):
+        self.path_to_ckpt = path_to_ckpt
+
+        self.detection_graph = tf.Graph()
+        with self.detection_graph.as_default():
+            od_graph_def = tf.compat.v1.GraphDef()
+            with tf.io.gfile.GFile(self.path_to_ckpt, 'rb') as fid:
+                serialized_graph = fid.read()
+                od_graph_def.ParseFromString(serialized_graph)
+                tf.import_graph_def(od_graph_def, name='')
+
+        self.default_graph = self.detection_graph.as_default()
+        self.sess = tf.compat.v1.Session(graph=self.detection_graph)
+
+        # Definite input and output Tensors for detection_graph
+        self.image_tensor = self.detection_graph.get_tensor_by_name(
+            'image_tensor:0')
+        # Each box represents a part of the image where a particular object was detected.
+        self.detection_boxes = self.detection_graph.get_tensor_by_name(
+            'detection_boxes:0')
+        # Each score represent how level of confidence for each of the objects.
+        # Score is shown on the result image, together with the class label.
+        self.detection_scores = self.detection_graph.get_tensor_by_name(
+            'detection_scores:0')
+        self.detection_classes = self.detection_graph.get_tensor_by_name(
+            'detection_classes:0')
+        self.num_detections = self.detection_graph.get_tensor_by_name(
+            'num_detections:0')
+
+    def processFrame(self, image):
+        # Expand dimensions since the trained_model expects images to have shape: [1, None, None, 3]
+        image_np_expanded = np.expand_dims(image, axis=0)
+        # Actual detection.
+        start_time = time.time()
+        (boxes, scores, classes,
+         num) = self.sess.run([
+             self.detection_boxes, self.detection_scores,
+             self.detection_classes, self.num_detections
+         ],
+            feed_dict={self.image_tensor: image_np_expanded})
+        end_time = time.time()
+
+        print("Elapsed Time:", end_time - start_time)
+
+        im_height, im_width, _ = image.shape
+        boxes_list = [None for i in range(boxes.shape[1])]
+        for i in range(boxes.shape[1]):
+            boxes_list[i] = (int(boxes[0, i, 1] * im_width),
+                             int(boxes[0, i, 0] * im_height),
+                             int(boxes[0, i, 3] * im_width),
+                             int(boxes[0, i, 2] * im_height))
+
+        return boxes_list, scores[0].tolist(), [
+            int(x) for x in classes[0].tolist()
+        ], int(num[0])
+
+    def close(self):
+        self.sess.close()
+        self.default_graph.close()

examples/mask-detector-workflow/blurry-faces/src/auto_blur_image.py

@@ -0,0 +1,104 @@
+# author: Asmaa Mirkhan ~ 2019
+
+import os
+import argparse
+import cv2 as cv
+from DetectorAPI import DetectorAPI
+
+def blurBoxes(image, boxes):
+    """
+    Argument:
+    image -- the image that will be edited as a matrix
+    boxes -- list of boxes that will be blurred, each box must be int the format (x_top_left, y_top_left, x_bottom_right, y_bottom_right)
+    
+    Returns:
+    image -- the blurred image as a matrix
+    """
+
+    for box in boxes:
+        # unpack each box
+        x1, y1, x2, y2 = [d for d in box]
+
+        # crop the image due to the current box
+        sub = image[y1:y2, x1:x2]
+
+        # apply GaussianBlur on cropped area
+        blur = cv.blur(sub, (10, 10))
+
+        # paste blurred image on the original image
+        image[y1:y2, x1:x2] = blur
+
+    return image
+
+
+def main(args):
+    # assign model path and threshold
+    model_path = args.model_path
+    threshold = args.threshold
+
+    # create detection object
+    odapi = DetectorAPI(path_to_ckpt=model_path)
+
+    # open image
+    image = cv.imread(args.input_image)
+
+    # real face detection
+    boxes, scores, classes, num = odapi.processFrame(image)
+
+    # filter boxes due to threshold
+    # boxes are in (x_top_left, y_top_left, x_bottom_right, y_bottom_right) format
+    boxes = [boxes[i] for i in range(0, num) if scores[i] > threshold]
+
+    # apply blurring
+    image = blurBoxes(image, boxes)
+
+    # # show image
+    # cv.imshow('blurred', image)
+
+    # if image will be saved then save it
+    if args.output_image:
+        cv.imwrite(args.output_image, image)
+        print('Image has been saved successfully at', args.output_image,
+              'path')
+    else:
+        cv.imshow('blurred', image)
+        # when any key has been pressed then close window and stop the program
+        cv.waitKey(0)
+        cv.destroyAllWindows()
+
+
+if __name__ == "__main__":
+    # creating argument parser
+    parser = argparse.ArgumentParser(description='Image blurring parameters')
+
+    # adding arguments
+    parser.add_argument('-i',
+                        '--input_image',
+                        help='Path to your image',
+                        type=str,
+                        required=True)
+    parser.add_argument('-m',
+                        '--model_path',
+                        default='/opt/blurry-faces/face_model/face.pb',
+                        help='Path to .pb model',
+                        type=str)
+    parser.add_argument('-o',
+                        '--output_image',
+                        help='Output file path',
+                        type=str)
+    parser.add_argument('-t',
+                        '--threshold',
+                        help='Face detection confidence',
+                        default=0.7,
+                        type=float)
+    args = parser.parse_args()
+    print(args)
+    # if input image path is invalid then stop
+    assert os.path.isfile(args.input_image), 'Invalid input file'
+
+    # if output directory is invalid then stop
+    if args.output_image:
+        assert os.path.isdir(os.path.dirname(
+            args.output_image)), 'No such directory'
+
+    main(args)

examples/mask-detector-workflow/blurry-faces/src/auto_blur_video.py

@@ -0,0 +1,125 @@
+# author: Asmaa Mirkhan ~ 2019
+
+import os
+import argparse
+import cv2 as cv
+from DetectorAPI import DetectorAPI
+
+def blurBoxes(image, boxes):
+    """
+    Argument:
+    image -- the image that will be edited as a matrix
+    boxes -- list of boxes that will be blurred, each box must be int the format (x_top_left, y_top_left, x_bottom_right, y_bottom_right)
+
+    Returns:
+    image -- the blurred image as a matrix
+    """
+
+    for box in boxes:
+        # unpack each box
+        x1, y1, x2, y2 = [d for d in box]
+
+        # crop the image due to the current box
+        sub = image[y1:y2, x1:x2]
+
+        # apply GaussianBlur on cropped area
+        blur = cv.blur(sub, (25, 25))
+
+        # paste blurred image on the original image
+        image[y1:y2, x1:x2] = blur
+
+    return image
+
+
+def main(args):
+    # assign model path and threshold
+    model_path = args.model_path
+    threshold = args.threshold
+
+    # create detection object
+    odapi = DetectorAPI(path_to_ckpt=model_path)
+
+    # open video
+    capture = cv.VideoCapture(args.input_video)
+
+    # video width = capture.get(3)
+    # video height = capture.get(4)
+    # video fps = capture.get(5)
+
+    if args.output_video:
+        fourcc = cv.VideoWriter_fourcc(*'mp4v')
+        output = cv.VideoWriter(args.output_video, fourcc,
+                                20.0, (int(capture.get(3)), int(capture.get(4))))
+
+    frame_counter = 0
+    while True:
+        # read frame by frame
+        r, frame = capture.read()
+        frame_counter += 1
+
+        # the end of the video?
+        if frame is None:
+            break
+
+        key = cv.waitKey(1)
+        if key & 0xFF == ord('q'):
+            break
+        # real face detection
+        boxes, scores, classes, num = odapi.processFrame(frame)
+
+        # filter boxes due to threshold
+        # boxes are in (x_top_left, y_top_left, x_bottom_right, y_bottom_right) format
+        boxes = [boxes[i] for i in range(0, num) if scores[i] > threshold]
+
+        # apply blurring
+        frame = blurBoxes(frame, boxes)
+
+        # show image
+        cv.imshow('blurred', frame)
+
+    # if image will be saved then save it
+        if args.output_video:
+            output.write(frame)
+            print('Blurred video has been saved successfully at',
+                  args.output_video, 'path')
+
+    # when any key has been pressed then close window and stop the program
+
+    cv.destroyAllWindows()
+
+
+if __name__ == "__main__":
+    # creating argument parser
+    parser = argparse.ArgumentParser(description='Image blurring parameters')
+
+    # adding arguments
+    parser.add_argument('-i',
+                        '--input_video',
+                        help='Path to your video',
+                        type=str,
+                        required=True)
+    parser.add_argument('-m',
+                        '--model_path',
+                        help='Path to .pb model',
+                        type=str,
+                        required=True)
+    parser.add_argument('-o',
+                        '--output_video',
+                        help='Output file path',
+                        type=str)
+    parser.add_argument('-t',
+                        '--threshold',
+                        help='Face detection confidence',
+                        default=0.7,
+                        type=float)
+    args = parser.parse_args()
+    print(args)
+    # if input image path is invalid then stop
+    assert os.path.isfile(args.input_video), 'Invalid input file'
+
+    # if output directory is invalid then stop
+    if args.output_video:
+        assert os.path.isdir(os.path.dirname(
+            args.output_video)), 'No such directory'
+
+    main(args)