|
|
|
@ -1,89 +1,78 @@ |
|
|
|
|
import numpy as np |
|
|
|
|
import cv2 |
|
|
|
|
import itertools as it |
|
|
|
|
|
|
|
|
|
''' |
|
|
|
|
from scipy.io import loadmat |
|
|
|
|
Neural network digit recognition sample. |
|
|
|
|
Usage: |
|
|
|
|
digits.py |
|
|
|
|
|
|
|
|
|
m = loadmat('ex4data1.mat') |
|
|
|
|
X = m['X'].reshape(-1, 20, 20) |
|
|
|
|
X = np.transpose(X, (0, 2, 1)) |
|
|
|
|
img = np.vstack(map(np.hstack, X.reshape(-1, 100, 20, 20))) |
|
|
|
|
img = np.uint8(np.clip(img, 0, 1)*255) |
|
|
|
|
cv2.imwrite('digits.png', img) |
|
|
|
|
Sample loads a dataset of handwritten digits from 'digits.png'. |
|
|
|
|
Then it trains a neural network classifier on it and evaluates |
|
|
|
|
its classification accuracy. |
|
|
|
|
''' |
|
|
|
|
|
|
|
|
|
import numpy as np |
|
|
|
|
import cv2 |
|
|
|
|
from common import mosaic |
|
|
|
|
|
|
|
|
|
def unroll_responses(responses, class_n): |
|
|
|
|
'''[1, 0, 2, ...] -> [[0, 1, 0], [1, 0, 0], [0, 0, 1], ...]''' |
|
|
|
|
sample_n = len(responses) |
|
|
|
|
new_responses = np.zeros((sample_n, class_n), np.float32) |
|
|
|
|
new_responses[np.arange(sample_n), responses] = 1 |
|
|
|
|
return new_responses |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
SZ = 20 |
|
|
|
|
SZ = 20 # size of each digit is SZ x SZ |
|
|
|
|
CLASS_N = 10 |
|
|
|
|
digits_img = cv2.imread('digits.png', 0) |
|
|
|
|
|
|
|
|
|
# prepare dataset |
|
|
|
|
h, w = digits_img.shape |
|
|
|
|
digits = [np.hsplit(row, w/SZ) for row in np.vsplit(digits_img, h/SZ)] |
|
|
|
|
digits = np.float32(digits).reshape(-1, SZ*SZ) |
|
|
|
|
N = len(digits) |
|
|
|
|
labels = np.repeat(np.arange(10), N/10) |
|
|
|
|
labels = np.repeat(np.arange(CLASS_N), N/CLASS_N) |
|
|
|
|
|
|
|
|
|
# split it onto train and test subsets |
|
|
|
|
shuffle = np.random.permutation(N) |
|
|
|
|
train_n = int(0.9*N) |
|
|
|
|
|
|
|
|
|
digits_train, digits_test = np.split(digits[shuffle], [train_n]) |
|
|
|
|
labels_train, labels_test = np.split(labels[shuffle], [train_n]) |
|
|
|
|
|
|
|
|
|
labels_train_unrolled = unroll_responses(labels_train, 10) |
|
|
|
|
|
|
|
|
|
# train model |
|
|
|
|
model = cv2.ANN_MLP() |
|
|
|
|
layer_sizes = np.int32([SZ*SZ, 25, 10]) |
|
|
|
|
layer_sizes = np.int32([SZ*SZ, 25, CLASS_N]) |
|
|
|
|
model.create(layer_sizes) |
|
|
|
|
|
|
|
|
|
# CvANN_MLP_TrainParams::BACKPROP,0.001 |
|
|
|
|
params = dict( term_crit = (cv2.TERM_CRITERIA_COUNT, 300, 0.01), |
|
|
|
|
params = dict( term_crit = (cv2.TERM_CRITERIA_COUNT, 100, 0.01), |
|
|
|
|
train_method = cv2.ANN_MLP_TRAIN_PARAMS_BACKPROP, |
|
|
|
|
bp_dw_scale = 0.001, |
|
|
|
|
bp_moment_scale = 0.0 ) |
|
|
|
|
print 'training...' |
|
|
|
|
labels_train_unrolled = unroll_responses(labels_train, CLASS_N) |
|
|
|
|
model.train(digits_train, labels_train_unrolled, None, params=params) |
|
|
|
|
model.save('dig_nn.dat') |
|
|
|
|
model.load('dig_nn.dat') |
|
|
|
|
|
|
|
|
|
ret, resp = model.predict(digits_test) |
|
|
|
|
resp = resp.argmax(-1) |
|
|
|
|
error_mask = (resp == labels_test) |
|
|
|
|
print error_mask.mean() |
|
|
|
|
|
|
|
|
|
def grouper(n, iterable, fillvalue=None): |
|
|
|
|
"grouper(3, 'ABCDEFG', 'x') --> ABC DEF Gxx" |
|
|
|
|
args = [iter(iterable)] * n |
|
|
|
|
return it.izip_longest(fillvalue=fillvalue, *args) |
|
|
|
|
|
|
|
|
|
def mosaic(w, imgs): |
|
|
|
|
imgs = iter(imgs) |
|
|
|
|
img0 = imgs.next() |
|
|
|
|
pad = np.zeros_like(img0) |
|
|
|
|
imgs = it.chain([img0], imgs) |
|
|
|
|
rows = grouper(w, imgs, pad) |
|
|
|
|
return np.vstack(map(np.hstack, rows)) |
|
|
|
|
|
|
|
|
|
test_img = np.uint8(digits_test).reshape(-1, SZ, SZ) |
|
|
|
|
|
|
|
|
|
def vis_resp(img, flag): |
|
|
|
|
def evaluate(model, samples, labels): |
|
|
|
|
'''Evaluates classifier preformance on a given labeled samples set.''' |
|
|
|
|
ret, resp = model.predict(samples) |
|
|
|
|
resp = resp.argmax(-1) |
|
|
|
|
error_mask = (resp == labels) |
|
|
|
|
accuracy = error_mask.mean() |
|
|
|
|
return accuracy, error_mask |
|
|
|
|
|
|
|
|
|
# evaluate model |
|
|
|
|
train_accuracy, _ = evaluate(model, digits_train, labels_train) |
|
|
|
|
print 'train accuracy: ', train_accuracy |
|
|
|
|
test_accuracy, test_error_mask = evaluate(model, digits_test, labels_test) |
|
|
|
|
print 'test accuracy: ', test_accuracy |
|
|
|
|
|
|
|
|
|
# visualize test results |
|
|
|
|
vis = [] |
|
|
|
|
for img, flag in zip(digits_test, test_error_mask): |
|
|
|
|
img = np.uint8(img).reshape(SZ, SZ) |
|
|
|
|
img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR) |
|
|
|
|
if not flag: |
|
|
|
|
img[...,:2] = 0 |
|
|
|
|
return img |
|
|
|
|
|
|
|
|
|
test_img = mosaic(25, it.starmap(vis_resp, it.izip(test_img, error_mask))) |
|
|
|
|
cv2.imshow('test', test_img) |
|
|
|
|
vis.append(img) |
|
|
|
|
vis = mosaic(25, vis) |
|
|
|
|
cv2.imshow('test', vis) |
|
|
|
|
cv2.waitKey() |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|