Inspired by Lucid, Luna is a Feature Visualization package for Tensorflow2.
Project description
Luna
Inspired by Lucid, Luna is a Feature Visualization package for Tensorflow. While Lucid is not supporting Tensorflow 2, and thus misses out on many modern features and the ability to use modern models, Luna was built with Tensorflow 2 at its core.
Luna is under active development. It is not production-ready research code and misses many features of Lucid.
Usage
You can use this package directly with your code. If you place this package next to your python file, you can use it like this in your file:
from luna.pretrained_models import models
from luna.featurevis import featurevis, images, image_reader
model = models.get_model("inceptionV3")
# model.summary()
image = images.initialize_image(224, 224)
opt_param = featurevis.OptimizationParameters(iterations=2000, learning_rate=0.7)
aug_param = featurevis.AugmentationParameters(blur= False, scale= True, pad_crop=True, flip=False,
rotation=False, noise=False, color_aug=False)
loss, image = featurevis.visualize_filter(image, model, layer_name,
channel_num, opt_param, aug_param)
images.save_image(image, name="test")
image_reader.save_npy_as_png("test.npy")
Recomended Reading
- Feature Visualization
- The Building Blocks of Interpretability
- Using Artifcial Intelligence to Augment Human Intelligence
- Visualizing Representations: Deep Learning and Human Beings
- Differentiable Image Parameterizations
- Activation Atlas
Related Talks
- Lessons from a year of Distill ML Research (Shan Carter, OpenVisConf)
- Machine Learning for Visualization (Ian Johnson, OpenVisConf)
Community
While we admire their work, we have no affiliation with the Lucid authors or project. Nonetheless, if you are interested in research like this, the Distill slack (join link) might be a good place for you to get to know other people in this area.
On the awesome Distill slack, Lucid has its own #proj-lucid
channel, where general questions about the technology are discussed.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for luna_fviz-0.0.3-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 24cc829144f88555be54da572bf0447f3371ea4a5cd8f625803834965a48621d |
|
MD5 | 5ebe1236241a6a2c4c5d5d2d93ca5c8c |
|
BLAKE2b-256 | d210813b1e02a2ab10cba93ed72f92ca6808c73bd2b3f88964a883ae72196d71 |