def evaluate(self, batch_size):
if self.train_generator.n_validation > 0:
keypoint_generator = self.train_generator(n_outputs=1,
batch_size=batch_size,
validation=True,
confidence=False)
elif self.train_generator.n_validation == 0:
warnings.warn(
"`n_validation` is 0, so the training set will be used for model evaluation"
)
keypoint_generator = self.train_generator(n_outputs=1,
batch_size=batch_size,
validation=False,
confidence=False)
y_pred_list = []
confidence_list = []
y_error_list = []
euclidean_list = []
for idx in range(len(keypoint_generator)):
X, y_true = keypoint_generator[idx]
y_pred = self.predict_model.predict_on_batch(X)
confidence_list.append(y_pred[..., -1])
y_pred_coords = y_pred[..., :2]
y_pred_list.append(y_pred_coords)
errors = keypoint_errors(y_true, y_pred_coords)
y_error, euclidean, mae, mse, rmse = errors
y_error_list.append(y_error)
euclidean_list.append(euclidean)
y_pred = np.concatenate(y_pred_list)
confidence = np.concatenate(confidence_list)
y_error = np.concatenate(y_error_list)
euclidean = np.concatenate(euclidean_list)
evaluation_dict = {
"y_pred": y_pred,
"y_error": y_error,
"euclidean": euclidean,
"confidence": confidence,
}
return evaluation_dict
def evaluate(self, batch_size):
if self.data_generator.n_validation > 0:
keypoint_generator = self.data_generator(n_outputs=1,
batch_size=batch_size,
validation=True,
confidence=False)
elif self.data_generator.n_validation == 0:
warnings.warn(
"`n_validation` is 0, so the training set will be used for model evaluation"
)
keypoint_generator = self.data_generator(n_outputs=1,
batch_size=batch_size,
validation=False,
confidence=False)
metrics = []
keypoints = []
for idx in range(len(keypoint_generator)):
X, y_true = keypoint_generator[idx]
y_pred = self.predict_model.predict_on_batch(X)
y_pred = y_pred[..., :2]
errors = keypoint_errors(y_true, y_pred)
y_error, euclidean, mae, mse, rmse = errors
confidence = y_pred[..., -1]
metrics.append([euclidean, mae, mse, rmse, confidence])
keypoints.append([y_pred, y_error])
metrics = np.hstack(metrics)
keypoints = np.hstack(keypoints)
euclidean, mae, mse, rmse, confidence = metrics
y_pred, y_error = keypoints
evaluation_dict = {
"y_pred": y_pred,
"y_error": y_error,
"euclidean": euclidean,
"mae": mae,
"mse": mse,
"rmse": rmse,
"confidence": confidence,
}
return evaluation_dict