def test_custom_img(self):
max_pixel_value = lambda img: img.max()
custom = lambda img: img.tolist()
img_interface = Interface(max_pixel_value, "image", "label", interpretation=custom)
result = img_interface.interpret([gradio.test_data.BASE64_IMAGE])[0]
expected_result = np.asarray(decode_base64_to_image(gradio.test_data.BASE64_IMAGE).convert('RGB')).tolist()
self.assertEqual(result, expected_result)
def rebuild(self, dir, data):
"""
Default rebuild method to decode a base64 image
"""
im = processing_utils.decode_base64_to_image(data)
timestamp = datetime.datetime.now()
filename = f'input_{timestamp.strftime("%Y-%m-%d-%H-%M-%S")}.png'
im.save(f'{dir}/{filename}', 'PNG')
return filename
def preprocess(self, x):
"""
Default preprocessing method for is to convert the picture to black and white and resize to be 48x48
"""
im = processing_utils.decode_base64_to_image(x)
im = im.convert('RGB')
im = processing_utils.resize_and_crop(
im, (self.image_width, self.image_height))
return np.array(im)
def rebuild(self, dir, data):
"""
Default rebuild method to decode a base64 image
"""
im = processing_utils.decode_base64_to_image(data)
timestamp = datetime.datetime.now()
filename = 'output_{}_{}.png'.format(self.label, timestamp.strftime("%Y-%m-%d-%H-%M-%S"))
im.save('{}/{}'.format(dir, filename), 'PNG')
return filename
def get_interpretation_scores(self, x, scores, masks):
x = processing_utils.decode_base64_to_image(x)
x = np.array(x)
output_scores = np.zeros((x.shape[0], x.shape[1]))
for score, mask in zip(scores, masks):
output_scores += score * mask
max_val, min_val = np.max(output_scores), np.min(output_scores)
if max_val > 0:
output_scores = (output_scores - min_val) / (max_val - min_val)
return output_scores.tolist()
def preprocess(self, x):
im = processing_utils.decode_base64_to_image(x)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
im = im.convert(self.image_mode)
if self.shape is not None:
im = processing_utils.resize_and_crop(
im, (self.shape[0], self.shape[1]))
if self.type == "pil":
return im
elif self.type == "numpy":
return np.array(im)
elif self.type == "file":
file_obj = tempfile.NamedTemporaryFile()
im.save(file_obj.name)
return file_obj
def get_interpretation_scores(self, x, neighbors, scores, masks):
"""
Returns:
(List[List[float]]): A 2D array representing the interpretation score of each pixel of the image.
"""
x = processing_utils.decode_base64_to_image(x)
x = np.array(x)
output_scores = np.zeros((x.shape[0], x.shape[1]))
for score, mask in zip(scores, masks):
output_scores += score * mask
max_val, min_val = np.max(output_scores), np.min(output_scores)
if max_val > 0:
output_scores = (output_scores - min_val) / (max_val - min_val)
return output_scores.tolist()
def get_interpretation_neighbors(self, x):
x = processing_utils.decode_base64_to_image(x)
if self.shape is not None:
x = processing_utils.resize_and_crop(x, self.shape)
image = np.array(x)
segments_slic = slic(image, self.interpretation_segments, compactness=10, sigma=1)
leave_one_out_tokens, masks = [], []
replace_color = np.mean(image, axis=(0, 1))
for (i, segVal) in enumerate(np.unique(segments_slic)):
mask = segments_slic == segVal
white_screen = np.copy(image)
white_screen[segments_slic == segVal] = replace_color
leave_one_out_tokens.append(
processing_utils.encode_array_to_base64(white_screen))
masks.append(mask)
return leave_one_out_tokens, {"masks": masks}, True
def preprocess(self, x):
"""
Default preprocessing method for the SketchPad is to convert the sketch to black and white and resize 28x28
"""
im_transparent = processing_utils.decode_base64_to_image(x)
# Create a white background for the alpha channel
im = PIL.Image.new("RGBA", im_transparent.size, "WHITE")
im.paste(im_transparent, (0, 0), im_transparent)
im = im.convert('L')
if self.invert_colors:
im = PIL.ImageOps.invert(im)
im = im.resize((self.image_width, self.image_height))
if self.flatten:
array = np.array(im).flatten().reshape(
1, self.image_width * self.image_height)
else:
array = np.array(im).flatten().reshape(1, self.image_width,
self.image_height)
return array
def preprocess(self, x):
im = processing_utils.decode_base64_to_image(x)
fmt = im.format
with warnings.catch_warnings():
warnings.simplefilter("ignore")
im = im.convert(self.image_mode)
if self.shape is not None:
im = processing_utils.resize_and_crop(im, self.shape)
if self.invert_colors:
im = PIL.ImageOps.invert(im)
if self.type == "pil":
return im
elif self.type == "numpy":
return np.array(im)
elif self.type == "file":
file_obj = tempfile.NamedTemporaryFile(suffix=("."+fmt.lower() if fmt is not None else ".png"))
im.save(file_obj.name)
return file_obj
else:
raise ValueError("Unknown type: " + str(self.type) + ". Please choose from: 'numpy', 'pil', 'file'.")
