def worker_to_visual_words(wind):
for j in range(math.ceil(len(all_imagenames) / num_cores)):
img_ind = j * num_cores + wind
if img_ind < len(all_imagenames):
img_name = all_imagenames[img_ind]
print('converting %d-th image %s to visual words' %
(img_ind, img_name))
image = cv.imread('../data/%s' % img_name)
image = cv.cvtColor(
image,
cv.COLOR_BGR2RGB) # convert the image from bgr to rgb
wordMap = get_visual_words(image, dictionary, filterBank)
pickle.dump(
wordMap,
open('../data/%s_%s.pkl' % (img_name[:-4], point_method),
'wb'))
def batch_to_visual_words (num_cores, point_method):
print ('using %d threads for getting visual words' % num_cores)
meta = pickle.load (open('../data/traintest.pkl', 'rb'))
all_imagenames = meta['all_imagenames']
# dictionary = pickle.load (open('dictionary%s.pkl' % point_method, 'rb'))
dictionary = pickle.load(open('/Users/justindulay/Downloads/scene_classification/python/randomWords.pkl' , 'rb'))
filterBank = create_filterbank()
# def worker_to_visual_words (wind):
for j in range(math.ceil(len(all_imagenames) / num_cores)):
# img_ind = j * num_cores + wind
img_ind = j
if img_ind < len(all_imagenames):
img_name = all_imagenames[img_ind]
print ('converting %d-th image %s to visual words' % (img_ind, img_name))
image = cv2.imread ('../data/%s' % img_name)
# should be OK in standard BGR format
wordMap = get_visual_words (image, dictionary, filterBank)
# print('here, the wordMap is', wordMap)
pickle.dump (wordMap, open('../data/%s_%s.pkl' % (img_name[:-4], point_method), 'wb'))
print('here we are going to call get_image_features on the wordMap')
histogram = get_image_features(wordMap, len(dictionary))
# workers = []
# for i in range(num_cores):
# workers.append (multiprocessing.Process(target=worker_to_visual_words, args=(i,)))
# for worker in workers:
# worker.start()
# for worker in workers:
# worker.join()
print ('batch to visual words done!')
# -----fill in your implementation here --------
meta = pickle.load(open('visionRandom.pkl', 'rb'))
trainHist = meta['trainFeatures']
trainLabels = meta['trainLabels']
filterBank = meta['filterBank']
randomWordsDictionary = meta['dictionary']
all_accuracies = []
for k in range(40):
testHist = []
predLabels = []
correct = 0
for i in range(len(test_imagenames)):
img_name = test_imagenames[i]
img = cv2.imread ('../data/%s' % img_name)
wordMap = get_visual_words(img, randomWordsDictionary, filterBank)
testFts = get_image_features(wordMap, len(randomWordsDictionary))
testHist.append(testFts)
# minDist = 100000000000
kminDist = [10000000000 for _ in range(k)]
predLabel = None
min_train_label_indices = []
for j in range(len(trainHist)):
dst = get_image_distance(trainHist[j], testFts, method='chi')
if len(kminDist) > 0:
if dst < max(kminDist):
# minDist = dst
kminDist.append(dst)
kminDist.remove(min(kminDist))