def get_training_set_classification(self, n, rstart, cstart, rend, cend):
training_data = []
for r in range(rstart, rend, 10):
if (r % 100 == 0):
print("Training sample",
int((r - rstart) * 100 / (rend - rstart)),
"percent complete")
for c1 in range(cstart, cend, 10):
(res, disp) = load_pfm.lookupL(r, c1, self.arr0, self.arr1)
corr_pixel = c1 - disp
if res == 2:
c3 = int(corr_pixel)
c4 = c3 + 1
training_data.append(
(1, self.build_Feature_Vector(r, c1, r, c3, n)))
training_data.append(
(1, self.build_Feature_Vector(r, c1, r, c4, n)))
for c2 in range(cstart, c1, 50):
if abs(c1 - c2 - disp) < 1.0:
training_data.append(
(1, self.build_Feature_Vector(r, c1, r, c2, n)))
else:
training_data.append(
(0, self.build_Feature_Vector(r, c1, r, c2, n)))
print("Training sample loaded")
return training_data
def get_training_set_regression(self, n):
training_data = []
for r in range(800, 1400, 10):
if (r % 100 == 0):
print("Training sample", int((r - 800) / 6),
"percent complete")
for c1 in range(800, 1400, 10):
(res, disp) = load_pfm.lookupL(r, c1, self.arr0, self.arr1)
corr_pixel = c1 - disp
if res == 2:
c3 = int(corr_pixel)
c4 = c3 + 1
training_data.append(
(disp, self.build_Feature_Vector(r, c1, r, c3, n)))
training_data.append(
(disp, self.build_Feature_Vector(r, c1, r, c4, n)))
# training_data.append((c1, c3, disp, self.build_Feature_Vector(r, c1, r, c3, n)))
# training_data.append((c1, c4, disp, self.build_Feature_Vector(r, c1, r, c4, n)))
for c2 in range(800, c1, 50):
training_data.append(
(disp, self.build_Feature_Vector(r, c1, r, c2, n)))
# training_data.append((c1, c2, disp, self.build_Feature_Vector(r,c1, r, c2, n)))
print("Training sample loaded")
return training_data
def getCandidates(self, r, c):
candidates = []
lef = self.lefts[r - rstart][c1 - cstart]
for dis in range(0, 300):
righ = self.rights[r - rstart][(cend - 1 - c) + dis]
feature_vector = np.abs(lef - righ)
(res, disp) = load_pfm.lookupL(r, c, self.arr0, self.arr1)
candidates.append((c, r, dis, feature_vector))
return candidates
def getCandidates(r, c):
candidates = []
lef = loader.lefts[r-rstart][c-cstart]
(res, disp) = load_pfm.lookupL(r, c, loader.arr0, loader.arr1)
if res == 2:
img3_arr[r-rstart,c-cstart] = disp
for dis in range(0,300):
righ = loader.rights[r-rstart][(cend - 1 - c) + dis]
feature_vector = np.abs(lef - righ)
#print(disp)
candidates.append((c, r, dis, feature_vector))
return candidates
pixels = loader.img_lookup(loader.Left_Pixels, c + cstart, r+rstart)
img2_arr[r,c] = 0.299 * pixels[0] + 0.587 * pixels[1] + 0.114*pixels[2]
pixels = loader.img_lookup(loader.Right_Pixels, c + cstart - best_disps[r,c], r + rstart)
img4_arr[r,c] = 0.299 * pixels[0] + 0.587 * pixels[1] + 0.114*pixels[2]
#calculate stats and create output images
total = 0
total_within1 = 0
total_within2 = 0
total_within5 = 0
meansquare = 0.0
for r in range(rend-rstart):
for c in range(cend-cstart):
(res, disp) = load_pfm.lookupL(r + rstart, c + cstart, loader.arr0, loader.arr1)
if res == 2:
err = abs(disp - img_arr[r,c])
if err <= 1:
total_within1 += 1
if err <= 2:
total_within2 += 1
if err <= 5:
total_within5 += 1
meansquare += err **2
total += 1
print("within 1:", total_within1 / total)
print("within 2:", total_within2 / total)
print("within 5:", total_within5 / total)
print("mean square error:", math.sqrt(meansquare / total))