def image_predict():
image_path = os.path.join(tmp_dir, session["image_id"] + ".png")
image = prepare_for_prediction(model, image_path)
label_onehot = model.predict(image)[0]
result = [float(x) for x in list(label_onehot)]
return jsonify(result)
def explain(self, model, image_path):
image = prepare_for_prediction(model, image_path, expand_dims=False)
label_onehot = model.predict(tf.expand_dims(image, 0))[0]
label = tf.math.argmax(label_onehot)
image = tf.cast(image, dtype=tf.float64)
explainer = lime_image.LimeImageExplainer()
explanation = explainer.explain_instance(image.numpy(),
model.predict,
top_labels=5,
hide_color=0,
num_samples=500)
temp, mask = explanation.get_image_and_mask(label.numpy(),
positive_only=True,
num_features=5,
hide_rest=True)
result_image = mark_boundaries(temp / 2 + 0.5, mask)
result_image = tf.image.convert_image_dtype(result_image,
dtype=tf.uint8,
saturate=True)
image_base46 = get_base64png(result_image)
result = {"label_id": float(label), "image": image_base46}
return result
def explain(self,
model: tf.keras.Model,
image_path: str,
general_settings: GeneralSettings,
tool_settings: dict = None) -> list:
"""
Explain prediction for given model and image. The resulting dictionary should contain keys:
- label_id: int
- image: base64 encoded image
Method is implemented for cases, when explaining is done one label after another.
Provide content of explain_one()!
Otherwise overwrite this method and you can ignore explain_one() method.
:param model: model whose prediction we would like to explain
:param image_path: image to be explained
:param general_settings: common setup containing settings for every tool (number of images etc.)
:param tool_settings: values for of a tool setup in the form of {parameter name: parameter value}
:return: dictionary with the explanation
"""
image = prepare_for_prediction(model, image_path)
onehot, labels = self._get_labels(model, image, general_settings)
tool_settings = tool_settings if tool_settings is not None else {}
results = map(
lambda label: {
"label_id": int(label),
"probability": float(onehot[label]),
"image": self.explain_one(model, image_path, label,
tool_settings),
},
labels,
)
return list(results)
def _explain(self, model, image_path, explain_class):
image = prepare_for_prediction(model, image_path)
label_onehot = model.predict(image)[0]
label = tf.math.argmax(label_onehot)
result_image = explain_class().explain((image.numpy(), None),
model,
class_index=label.numpy())
image_base46 = get_base64png(result_image)
result = {"label_id": int(label), "image": image_base46}
return result
def explain(self,
model: tf.keras.Model,
image_path: str,
general_settings: GeneralSettings,
tool_settings: dict = None) -> list:
image = prepare_for_prediction(model, image_path)
onehot, labels = self._get_labels(model, image, general_settings)
image = tf.cast(image, dtype=tf.float64)
explainer = lime_image.LimeImageExplainer()
explanation = explainer.explain_instance(
image.numpy()[0],
model.predict,
labels=labels,
batch_size=tool_settings["batch_size"],
num_features=tool_settings["num_features"],
num_samples=tool_settings["num_samples"],
)
results = []
for label in labels:
temp, mask = explanation.get_image_and_mask(
label,
positive_only=tool_settings["positive_only"],
num_features=tool_settings["superpixels"],
min_weight=tool_settings["min_weight"],
hide_rest=tool_settings["hide_rest"],
)
result_image = mark_boundaries(temp / 2 + 0.5, mask)
result_image = tf.image.convert_image_dtype(result_image,
dtype=tf.uint8,
saturate=True)
image_base46 = get_base64png(result_image)
results.append({
"label_id": int(label),
"probability": float(onehot[label]),
"image": image_base46,
})
return results
def explain_one(self, model, image_path, label, tool_settings):
image = prepare_for_prediction(model, image_path)
result_image = self.explain_class().explain((image.numpy(), None), model, class_index=label, **tool_settings)
image_base46 = get_base64png(result_image)
return image_base46