Cnn-Hypercolumn

Legacy repository. Preserved as older work from 2016-2017 and archived. It is not representative of my current research program.

Cnn-Hypercolumn is a small experiment around convolutional hypercolumns for image description and clustering.

What Is Here

• Cnn-Hypercolumn.py as the main script
• vgg-16_keras.py and vgg-19_keras.py as model/demo entry points
• VGG weight links for reproducing the original setup

Setup

Install the Python dependencies:

pip install -r requirements.txt

If you want an isolated environment:

python -m venv .env
source .env/bin/activate
pip install -r requirements.txt

References

• Background post: Convolutional Hypercolumns in Python
• VGG16 weights: vgg16_weights.h5
• VGG19 weights: vgg19_weights.h5

Notes

The clustering step at the end of the main script can be memory-intensive on small machines.