def InferForImageArray(self, image_numpy_arr):
if self._rescale_in_preprocessing:
image_numpy_arr = ImagePreprocessor.RescaleImageInput(
image_numpy_arr)
else:
image_numpy_arr = ImagePreprocessor.NormalizeImageInput(
image_numpy_arr)
inference_inputs = [numpy.expand_dims(image_numpy_arr, axis=0)]
if self._use_mask_input:
inference_inputs.append(numpy.ones((1, self._num_rnn_steps)))
predicted_char_id, predicted_probs = self._inference_fn(
*inference_inputs)
chars = self.captcha_model.CHARS
if predicted_char_id.ndim == 1:
predicted_chars = chars[predicted_char_id[0]]
probs_by_chars = {}
for i in range(predicted_probs.shape[1]):
probs_by_chars[chars[i]] = predicted_probs[0, i]
else:
assert predicted_char_id.ndim == 2, predicted_char_id.shape
predicted_chars = [0] * predicted_char_id.shape[1]
probs_by_chars = [{} for _ in range(predicted_char_id.shape[1])]
for i in range(predicted_char_id.shape[1]):
predicted_chars[i] = chars[predicted_char_id[0, i]]
for j in range(predicted_probs.shape[2]):
probs_by_chars[i][chars[j]] = predicted_probs[0, i, j]
return predicted_chars, probs_by_chars
def Load(cls, file_path, rescale_in_preprocessing=False):
training_data = numpy.load(file_path)
image_input = training_data['image_data']
if rescale_in_preprocessing:
for row in range(image_input.shape[0]):
image_input[row, 0, :, :] = ImagePreprocessor.RescaleImageInput(
image_input[row, 0, :, :])
else:
image_input = ImagePreprocessor.NormalizeImageInput(image_input)
ret = (image_input, training_data['chars'])
del training_data.f
training_data.close()
return ret
def read_and_parse(file_content,cracker):
im = read_data(file_content)
array = numpy.asarray(im.convert('L')).copy()
image_input = ImagePreprocessor.ProcessImage(array)
#print "saving image"
#im.save("geet.png")
predicted_chars, char_probabilities = cracker.InferForImageArray(image_input)
return "".join(x for x in predicted_chars)
def GenerateTrainingData(cls,
captchas_dir,
training_data_dir,
max_size=20000,
max_captcha_length=8):
image_shape = _GetShapeOfImagesUnderDir(captchas_dir)
training_data_shape = tuple(
[max_size] +
list(ImagePreprocessor.GetProcessedImageShape(image_shape)))
training_image_data = numpy.zeros(training_data_shape,
dtype=numpy.float32)
training_labels = numpy.zeros((max_size, max_captcha_length),
dtype=numpy.int32)
i = 0
for captcha_filepath in utils.GetFilePathsUnderDir(captchas_dir):
try:
image_data = ImagePreprocessor.GetImageData(captcha_filepath)
except Exception as e:
print e, captcha_filepath
continue
i += 1
index = i % max_size
training_image_data[index] = ImagePreprocessor.ProcessImage(
image_data)
captcha_ids = _GetCaptchaIdsFromImageFilename(captcha_filepath)
training_labels[index, :] = numpy.zeros(max_captcha_length,
dtype=numpy.int32)
training_labels[index, :captcha_ids.shape[0]] = captcha_ids
if i != 0 and (i % 10000) == 0:
print 'Generated {0} examples.'.format(i)
if i != 0 and i % max_size == 0:
file_path = os.path.join(
training_data_dir,
"training_images_{0}.npy".format(i / max_size))
try:
cls.Save(file_path, training_image_data, training_labels)
except Exception as e:
print e
def InferFromImagePath(self, image_path):
image_input = ImagePreprocessor.ProcessImage(
ImagePreprocessor.GetImageData(image_path))
print("image_input%%%%", image_input.shape)
return self.InferForImageArray(image_input.copy())
def _GetShapeOfImagesUnderDir(captchas_dir):
for captcha_filepath in utils.GetFilePathsUnderDir(captchas_dir):
image_data = ImagePreprocessor.GetImageData(captcha_filepath)
return image_data.shape
return None