Skip to content

MaskRCNN:Example:VideoReadWrite

Mask R-CNN 데모를 위한 Video Read & Write Python example.

Source code

실행을 위해 samples/coco/coco.py파일이 함께 필요하다. import os import sys import random import math import numpy as np import skimage.io import matplotlib import matplotlib.pyplot as plt

  1. Import Mask RCNN

from mrcnn import utils import mrcnn.model as modellib from mrcnn import visualize

  1. Root directory of the project

ROOT_DIR = os.getcwd()

  1. Import COCO config

sys.path.append(ROOT_DIR) import coco

  1. Directory to save logs and trained model

MODEL_DIR = ROOT_DIR

  1. Local path to trained weights file

COCO_MODEL_PATH = os.path.join(ROOT_DIR, "mask_rcnn_coco.h5")

  1. Download COCO trained weights from Releases if needed

if not os.path.exists(COCO_MODEL_PATH):

utils.download_trained_weights(COCO_MODEL_PATH)

class InferenceConfig(coco.CocoConfig):

# Set batch size to 1 since we'll be running inference on
# one image at a time. Batch size = GPU_COUNT * IMAGES_PER_GPU
GPU_COUNT = 1
IMAGES_PER_GPU = 1
DETECTION_MIN_CONFIDENCE = 0.6
#RPN_NMS_THRESHOLD = 0.5

config = InferenceConfig() config.display()

  1. Create model object in inference mode.

model = modellib.MaskRCNN(mode="inference", model_dir=MODEL_DIR, config=config)

  1. Load weights trained on MS-COCO

model.load_weights(COCO_MODEL_PATH, by_name=True)

  1. COCO Class names
  2. Index of the class in the list is its ID. For example, to get ID of
  3. the teddy bear class, use: class_names.index('teddy bear')

class_names = ['BG', 'person', 'bicycle', 'car', 'motorcycle', 'airplane',

'bus', 'train', 'truck', 'boat', 'traffic light',
'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird',
'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear',
'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie',
'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball',
'kite', 'baseball bat', 'baseball glove', 'skateboard',
'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup',
'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple',
'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza',
'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed',
'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote',
'keyboard', 'cell phone', 'microwave', 'oven', 'toaster',
'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors',
'teddy bear', 'hair drier', 'toothbrush']

  1. car: 3
  2. bus: 6
  3. truck: 8
  1. Load a random image from the images folder
  2. image = skimage.io.imread(os.path.join(ROOT_DIR, 'zz.jpg'))
  1. Run detection
  2. results = model.detect([image], verbose=1)
  1. Visualize results
  2. r = results[0]
  3. visualize.display_instances(image, r['rois'], r['masks'], r['class_ids'], class_names, r['scores'])

import math

def drawImage(image, boxes, masks, class_ids, class_names, scores, show_bbox=False, show_label=False, show_seg=True, show_center=True):

N = boxes.shape[0] # Number of instances
if not N:
print("\n*** No instances to display *** \n")
else:
assert boxes.shape[0] == masks.shape[-1] == class_ids.shape[0]

masked_image = image.astype(np.uint32).copy()

car_count = 0
for i in range(N):
class_id = class_ids[i]
if class_id != 3 and class_id != 6 and class_id != 8:
continue

car_count = car_count + 1

# Bounding Box.
if not np.any(boxes[i]):
# Skip this instance. Has no bbox. Likely lost in image cropping.
continue
y1, x1, y2, x2 = boxes[i]

if show_bbox:
cv2.rectangle(image, (x1, y1), (x2, y2), (0, 255, 0), 1, cv2.LINE_AA)
#p = patches.Rectangle((x1, y1), x2 - x1, y2 - y1, linewidth=2, alpha=0.7, linestyle="dashed", edgecolor=color, facecolor='none')

if show_center:
cv2.circle(image, (x1 + int(math.fabs(float(x2 - x1))/2), y1 + int(math.fabs(float(y2 - y1))/2)), 4, (0, 0, 255), -1)

if show_label:
score = scores[i] if scores is not None else None
label = class_names[class_id]
caption = "{} {:.3f}".format(label, score) if score else label

font = cv2.FONT_HERSHEY_SIMPLEX
cv2.putText(image, caption, (x1, y1 + 8), font, 1, (255, 255, 255), 1, cv2.LINE_AA)
#ax.text(x1, y1 + 8, caption, color='w', size=11, backgroundcolor="none")

if show_seg:
# Mask
mask = masks[:, :, i]
masked_image = visualize.apply_mask(masked_image, mask, (0, 0, 0))

# Mask Polyline
# Pad to ensure proper polygons for masks that touch image edges.
padded_mask = np.zeros((mask.shape[0] + 2, mask.shape[1] + 2), dtype=np.uint8)
padded_mask[1:-1, 1:-1] = mask
contours = visualize.find_contours(padded_mask, 0.5)
for verts in contours:
# Subtract the padding and flip (y, x) to (x, y)
verts = np.fliplr(verts) - 1
cv2.polylines(image, np.int32([verts]), True, (0, 255, 255))
#p = visualize.Polygon(verts, facecolor="none", edgecolor=())
#ax.add_patch(p)

font = cv2.FONT_HERSHEY_SIMPLEX
cv2.putText(image, 'CAR: {}'.format(car_count), (5, 32), font, 1, (0, 0, 255), 2, cv2.LINE_AA)
pass

import numpy as np import cv2

INPUT_VIDEO_PATH = os.path.join(ROOT_DIR, 'video1.mp4')

cap = cv2.VideoCapture(INPUT_VIDEO_PATH) VIDEO_WIDTH = cap.get(cv2.CAP_PROP_FRAME_WIDTH) VIDEO_HEIGHT = cap.get(cv2.CAP_PROP_FRAME_HEIGHT) VIDEO_FRAMES = cap.get(cv2.CAP_PROP_FRAME_COUNT) VIDEO_FPS = cap.get(cv2.CAP_PROP_FPS)

print('Video Open: ', INPUT_VIDEO_PATH) print('Video size: ', VIDEO_WIDTH, 'x', VIDEO_HEIGHT) print('Video frames: ', VIDEO_FRAMES) print('Video FPS: ', VIDEO_FPS)

FOURCC = cv2.VideoWriter_fourcc(*'XVID') out = cv2.VideoWriter('output.avi', FOURCC, float(VIDEO_FPS), (int(VIDEO_WIDTH), int(VIDEO_HEIGHT)))

frame_index = 0 SKIP_PREFIX_FRAMES = 0

show_video = True

while (cap.isOpened()):

ret, frame = cap.read()

if (frame_index < SKIP_PREFIX_FRAMES):
frame_index = frame_index + 1
continue

results = model.detect([frame], verbose=1)
print('Detect: ', frame_index, '/', VIDEO_FRAMES)
frame_index = frame_index + 1

r = results[0]
drawImage(frame, r['rois'], r['masks'], r['class_ids'], class_names, r['scores'])

if show_video:
cv2.imshow('preview', frame)
out.write(frame)

key_code = cv2.waitKey(1)
if key_code & 0xFF == ord('q'):
break
elif key_code & 0xFF == ord('s'):
show_video = True
elif key_code & 0xFF == ord('h'):
show_video = False

cap.release() out.release() cv2.destroyAllWindows()

</syntaxhighlight>

See also

Favorite site