#Keras data augmentation github code#
In my previous blog post Achieving Top 23% in Kaggle's Facial Keypoints Detection with Keras + Tensorflow, I implemented a python class that can flip the image horizontally and shift the image both along horizontal and vertical axes while adjusting the landmark coordinates.īut there are so many other translations that I want to do e.g., shearing, zooming, or all of them at once! And I do not want to code rotation matrix by myself! Keras's ImageDataGenerator for facial keypoint detection problem. I was looking for some existing API that can translate both images and coordinates. That means that if the image of a face is shifted by 3 pixels, the (x,y) coordinates of the eye location also needs to be shifted. In landmark detection or facial keypoint detections, the target values also needs to change when an image is translated. So the target label "dog" does not need to be translated. For example, the image of a dog is still an image of a dog even if the image is shifted by 3 pixels.
#Keras data augmentation github generator#
This generator has been used in many of my previous blog posts, for example: The generator can generate augmented images from the training images on the fly. Keras has a powerful API called ImageDataGenerator that resolve this problem. Keras's ImageDataGenerator and its limit ¶ĭata augmentation could increase the number of training images substantially which could raise a storage problem.
This experiment shows that it is essential to increase the data size using data augmentation to develop a robust deep learning model. However, the model performance improves when training data also contains translated images. In my previous blog post, I have seen poor performance of a deep learning model when testing images contain the translation of the training images. Why data augmentation? ¶ĭeep learning model is data greedy and the performance of the model may be surprisingly bad when testing images vary from training images a lot.ĭata augmentation is an essential technique to utilize limited amount of training images.
The python class ImageDataGenerator_landmarks is available at my github account.
0 kommentar(er)
