Category : data-augmentation

I’m now running Wide ResNet on CIFAR dataset (CIFAR-10 and CIFAR-100). One confusing thing for me is that if I did data pre-processing (random horizontally flip and random crop), the test accuracy is much better than the accuracy that I used just normalized raw data. The code of pre-processing is shown below: def train_prep(x, y): ..

Read more

For the training of my model I want to perform data augmentation on the data set to improve performance. My input data instances consist of snapshots of wave height, saved in CSV (a 160×160 grid), which is the same for my labels/reference data. For that reason I want to perform the same data augmentation procedure ..

Read more

Good day, I am trying to implement a github repo specAugment (https://github.com/DemisEom/SpecAugment) After loading the wav file using librosa, I believe it uses numPy reshape function to reshape the melspectrogram array, get Log scale melspectrogram by using power_to_db function and apply the data augmentation. My question is, is it possible to get a wav file ..

Read more

I am working on Semantic segmentation using U-net and I’m trying to augment training data using ImageDataGenerator. There is one parameter whose effect I don’t completely understand – the parameter rounds in the .fit part shown below in the code. I have checked the Keras documentation (https://www.tensorflow.org/api_docs/python/tf/keras/preprocessing/image/ImageDataGenerator#fit) and it says that rounds parameter does the ..

Read more

Preparing my data with ImageDataGenerator. So far I did the following, For training data: image_data_generator_training = tf.keras.preprocessing.image.ImageDataGenerator( **process** validation_split=0.2 ).flow_from_directory(‘./dataset/fg_image’, batch_size = 16, target_size = (224, 224), seed = SEED, subset=’training’) mask_data_generator_training = tf.keras.preprocessing.image.ImageDataGenerator( **process** validation_split=0.2 ).flow_from_directory(‘./dataset/gt_mask’, batch_size = 16, target_size = (224, 224), seed = SEED, subset=’training’) For validation data: image_data_generator_validation = tf.keras.preprocessing.image.ImageDataGenerator( **process** ..

Read more