Outsource the object recognition in a different process.

Author: datitran

README.md

@@ -2,7 +2,16 @@
 
 A simple example demonstrating [Google's TensorFlow Object Detection API](https://github.com/tensorflow/models/tree/master/object_detection) with [OpenCV](http://opencv.org/) to recognizes objects through a webcam.
 
+## Getting Started
+1. `conda env create -f environment.yml`
+2. `python object_detection_app.py`
+
+## Requirements
+- [Anaconda / Python 3.5](https://www.continuum.io/downloads)
+- [TensorFlow 1.2](https://www.tensorflow.org/)
+- [OpenCV 3.1](http://opencv.org/)
+
 ## Copyright
 
 See [LICENSE](LICENSE) for details.
-Copyright (c) 2017 [Dat Tran](http://www.dat-tran.com/).
+Copyright (c) 2017 [Dat Tran](http://www.dat-tran.com/).

environment.yml

@@ -0,0 +1,38 @@
+name: object-detection
+channels: !!python/tuple
+- menpo
+- defaults
+dependencies:
+- freetype=2.5.5=2
+- hdf5=1.8.17=1
+- jbig=2.1=0
+- jpeg=9b=0
+- libpng=1.6.27=0
+- libtiff=4.0.6=3
+- menpo::opencv3=3.1.0=py35_0
+- menpo::tbb=4.3_20141023=0
+- mkl=2017.0.1=0
+- numpy=1.13.0=py35_0
+- olefile=0.44=py35_0
+- openssl=1.0.2l=0
+- pillow=4.1.1=py35_0
+- pip=9.0.1=py35_1
+- python=3.5.3=1
+- readline=6.2=2
+- setuptools=27.2.0=py35_0
+- sqlite=3.13.0=0
+- tk=8.5.18=0
+- wheel=0.29.0=py35_0
+- xz=5.2.2=1
+- zlib=1.2.8=3
+- pip:
+ - backports.weakref==1.0rc1
+ - bleach==1.5.0
+ - html5lib==0.9999999
+ - markdown==2.2.0
+ - protobuf==3.3.0
+ - six==1.10.0
+ - tensorflow==1.2.0
+ - werkzeug==0.12.2
+prefix: /Users/datitran/anaconda/envs/object-detection
+

object_detection_app.py

@@ -1,10 +1,12 @@
 import time
 import os
 import cv2
+import multiprocessing
 import numpy as np
 import tensorflow as tf
 
 from utils import FPS, WebcamVideoStream
+from multiprocessing import Process, Queue, Pool
 from object_detection.utils import label_map_util
 from object_detection.utils import visualization_utils as vis_util
 
@@ -19,6 +21,9 @@
 
 NUM_CLASSES = 90
 
+NUM_WORKERS = 2 # cv2.getNumberOfCPUs() - 1
+QUEUE_SIZE = 5
+
 # Loading label map
 label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
 categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES,
@@ -57,7 +62,7 @@ def detect_objects(image_np, sess, detection_graph):
  return image_np
 
 
-if __name__ == '__main__':
+def worker(input_q, output_q):
  # Load a (frozen) Tensorflow model into memory.
  detection_graph = tf.Graph()
  with detection_graph.as_default():
@@ -69,34 +74,48 @@ def detect_objects(image_np, sess, detection_graph):
 
  sess = tf.Session(graph=detection_graph)
 
- video_capture = WebcamVideoStream(src=0).start()
  fps = FPS().start()
+ while True:
+ fps.update()
+ frame = input_q.get()
+ output_q.put(detect_objects(frame, sess, detection_graph))
+
+ fps.stop()
+ sess.close()
 
- while fps._numFrames < 120:
- frame = video_capture.read()
 
- t = time.time()
+if __name__ == '__main__':
+ logger = multiprocessing.log_to_stderr()
+ logger.setLevel(multiprocessing.SUBDEBUG)
+
+ input_q = Queue(maxsize=QUEUE_SIZE)
+ output_q = Queue(maxsize=QUEUE_SIZE)
+
+ process = Process(target=worker, args=((input_q, output_q)))
+ process.daemon = True
+ pool = Pool(NUM_WORKERS, worker, (input_q, output_q))
+
+ video_capture = WebcamVideoStream(src=0).start()
+ fps = FPS().start()
 
- # cv2.imshow('Video', detect_objects(frame, sess, detection_graph))
+ while True: # fps._numFrames < 120
+ frame = video_capture.read()
+ input_q.put(frame)
 
- # time.sleep(2)
+ # t = time.time()
 
- detect_objects(frame, sess, detection_graph)
+ cv2.imshow('Video', output_q.get())
 
- print(time.time() - t)
+ # print(time.time() - t)
 
- # update the FPS counter
  fps.update()
 
  if cv2.waitKey(1) & 0xFF == ord('q'):
  break
 
- # stop the timer and display FPS information
  fps.stop()
  print('[INFO] elasped time: {:.2f}'.format(fps.elapsed()))
  print('[INFO] approx. FPS: {:.2f}'.format(fps.fps()))
 
- # When everything is done, release the capture
  video_capture.stop()
  cv2.destroyAllWindows()
- sess.close()

utils.py

@@ -1,3 +1,5 @@
+# From http://www.pyimagesearch.com/2015/12/21/increasing-webcam-fps-with-python-and-opencv/
+
 import cv2
 import datetime
 from threading import Thread