Juggle

Inspired by Adrian Rosebrock Pyimagesearch starter, practitioner, and imagenet bundle DL4CV.

pip install juggle

The juggle directory contains the classic pyimagesearch tools with additional deep learnig tools

Differences with pyimagesearch

These are the 5 following changes:

• The first letter of every word in a file's title has been switched to uppercase, e.g.,simpledatasetloader.py is now SimpleDatasetLoader.py
• All Keras libraries have been already moved to tensorflow.keras so is up to date with TensorFlow 2.0 ,i.e,

# Before
from keras.preprocessing.image import load_img
# Now
from tensorflow.keras.preprocessing.image import load_img

• SimpleDatasetLoader.py previously located in pyimageserach/datasets has been moved to preprocessing to match Keras style

from tensorflow.keras.preprocessing.image import load_img
from juggle.preprocessing import SimpleDatasetLoader

• All training sample codes with plots have been modified to meet (TF 2.0 Metrics [acc] vs [accuracy])[https://github.com/keras-team/keras/releases/tag/2.3.0]. All metrics have been switched to [acc] for backward compatibility. In summary the new changes suggest that if you specify metrics=["accuracy"] in the model.compile(), then the history object will have the keys as 'accuracy' and 'val_accuracy'. While if you specify it as metrics=["acc"] then they will be reported with the keys 'acc' and 'val_acc'.
• General TF 2.0 Differences

# BEFORE
from keras.utils.np_utils import to_categorical
from keras.layers.core import Dense
from keras.layers.convolutional import Conv2D

# NOW
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.layers import Dense, Conv2D