https://github.com/phtrempe/netception
A neural network inception library
https://github.com/phtrempe/netception
Last synced: 3 months ago
JSON representation
A neural network inception library
- Host: GitHub
- URL: https://github.com/phtrempe/netception
- Owner: PhTrempe
- License: mit
- Created: 2017-07-02T19:32:55.000Z (almost 8 years ago)
- Default Branch: master
- Last Pushed: 2017-07-04T00:12:38.000Z (almost 8 years ago)
- Last Synced: 2025-02-13T10:48:52.731Z (3 months ago)
- Language: Python
- Size: 13.7 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Netception
## Description
Netception is a neural network inception library developed
in Python 3 and used with Keras.
## How to Install
```
pip install netception
```
## Dependencies
Here are Netception's required dependencies.
* `numpy`
* `keras`
Here are some optional dependencies one might need.
* `h5py` (to load and save Keras models stored in files)
* `pillow` (to manipulate images)
Also note that Keras requires a backend library like TensorFlow to operate.
```
pip install tensorflow
```
If GPU support is desired, it is also possible to use the GPU version of
TensorFlow.
```
pip install tensorflow-gpu
```
## An Exhaustively Commented Example to Get You Started
```python
import os
from PIL import Image
from keras import applications, backend
from netception.inceptor import Inceptor
from netception.utils.visualization_util import VisualizationUtil
if __name__ == "__main__":
# Load the model to incept
# (Here, we load the pretrained VGG16 model from Keras)
model = applications.VGG16()
# Print the model's summary to see its layers
model.summary()
# Determine the target to incept within the model
# (Here, we choose to incept the output of the 455th filter of the
# convolutional layer "block5_conv3")
target = model.get_layer("block5_conv3").output[:, :, :, 455]
# Create an inceptor and configure it
# (Here, we create an inceptor with our model and target. We also set an
# inception rate of 0.25, a maximal number of steps of 50, and parameters
# for early stopping if the inception score stops improving enough)
inceptor = Inceptor(
model=model,
target=target,
inception_rate=0.5,
max_steps=200,
improvement_check_interval=5,
improvement_threshold=0.05
)
# Run the inceptor
inception, score = inceptor.incept()
# Convert the resulting inception into image data
image_data = VisualizationUtil.inception_to_bytes(
inception=inception,
colorfulness=0.15
)
# Create an image from the image data, and resize the image
image = Image.fromarray(image_data).resize((512, 512), Image.BICUBIC)
# Show the image
image.show()
# Save the image
script_dir = os.path.dirname(os.path.realpath(__file__))
image.save(os.path.join(script_dir, "inception.png"))
# Clear the backend session
backend.clear_session()
```
This is what the result looks like.
## The Same Example In Compact Form For a Quick Copy & Paste
```python
import os
from PIL import Image
from keras import applications, backend
from netception.inceptor import Inceptor
from netception.utils.visualization_util import VisualizationUtil
if __name__ == "__main__":
model = applications.VGG16()
model.summary()
target = model.get_layer("block5_conv3").output[:, :, :, 455]
inceptor = Inceptor(model, target, 0.5, 200, 5, 0.05)
inception, score = inceptor.incept()
image_data = VisualizationUtil.inception_to_bytes(inception, 0.15)
image = Image.fromarray(image_data).resize((512, 512), Image.BICUBIC)
image.show()
script_dir = os.path.dirname(os.path.realpath(__file__))
image.save(os.path.join(script_dir, "inception.png"))
backend.clear_session()
```