def detect():
''' Given a classifier, return a set of locations
from which we will draw on our final image
strong = (h, curr_fpr, curr_fnr, big_theta)
h = (alpha_t, h_t, featureIdx , polar, thresh)
'''
cascade = np.load("./data/cascades/cascade_2000.npy")
feat_index = np.load("./data/features/feat_index.npy")
test_idx, test_imgs = trim()
for strong in cascade:
h = strong[0]
big_theta = strong[3]
h_test = [] # all rounds of boosting
h_single = np.zeros(len(test_imgs)) # One round of boosting
for weak in h:
alpha_t = weak[0]
thresh = weak[4]
parity = weak[3]
featureIdx= weak[2]
feature = feat_index[featureIdx]
for i in range(len(test_imgs)):
img_gray = test_imgs[i]
h_single[i] = apply_weak_classifier(img_gray,thresh,parity,feature)
h_test.append(tuple( (alpha_t, h_single) ))
signed_test = calc_classifiers(h_test,big_theta)
# remove all the zeros in the signed_test
remove_idx = []
for j in range(len(signed_test)):
if signed_test[j] == 0 :
remove_idx.append(j)
# remove images and corresponding indexes
test_imgs = np.delete(test_imgs, remove_idx, axis = 0)
test_idx = np.delete(test_idx, remove_idx, axis = 0)
draw("out.png",test_idx)
def apply_validation(h,valid,feat_index,label,truth,big_theta):
''' CHANGED ZERO TO NEG
Input:
h = (alpha, h_t, featureIdx, polar, thresh) 2d array T entries
valid = dictionary containing all validation rectangles
feat_index = lookup table for all features
label = truth vector for all training images
Output:
Current false positive and false negative rates
Find big theta, and run the validation set and
training set on that
'''
pos_list = valid["pos"]
neg_list = valid["neg"]
combined = pos_list+neg_list # first half is positive, second is negative
h_valid = [] # all rounds of boosting
h_valid_temp = np.zeros(len(combined)) # One round of boosting
# create the shfited classifier
for t in range(len(h)):
alpha_t = h[t][0]
thresh = h[t][4]
parity = h[t][3]
feature = feat_index[h[t][2]]
# creates h_i
print "Applying the weak classificaiton to a validation data"
for i in range(len(combined)):
# for each image in combined, apply the weak classifier
img_gray = combined[i]
# classified array of 1,0, 800x1
h_valid_temp[i] = apply_weak_classifier(img_gray,thresh,parity,feature)
h_valid.append(tuple(( alpha_t, h_valid_temp )))
# print "MY BIG THETA IS :", big_theta
return test_big_theta(big_theta, h_valid, truth, label, h)
