UberNet: A Jupyter Notebook repository from takmin

About this folk

This repository is forked from https://github.com/jkokkin/UberNet

This code is modified from original code to run in Python3.

Intro

First release of UberNet in "test" mode. The demo performs all of the UberNet tasks using a VGG16-architecture network.

This repository builds on the py-faster-rcnn (https://github.com/rbgirshick/py-faster-rcnn) release of faster-rcnn, which in turn builds on the caffe library. The py-faster-rcnn instructions directly allow one to compile the present code (copied below for convenience). The main changes to the original py-faster-rcnn release have been annotated with a prefix and a suffix.

Forthcoming:

code that allows to get normalized cut eigenvectors from image boundaries.
training code.
ResNet-based models.

License

UberNet is released under the GPL License (refer to the UberNet-LICENSE file for details). Faster-RCNN is released under the MIT License (refer to the Faster-RCNN-LICENSE file for details).

Citing UberNet

If you find UberNet useful in your research, please consider citing: @inproceedings{ubernet, Author = {Iasonas Kokkinos}, Title = {UberNet: Training a `Universal' Convolutional Neural Network for Low-, Mid-, and High-Level Vision using Diverse Datasets and Limited Memory}, Booktitle = {Computer Vision and Pattern Recognition (CVPR)}, Year = {2017} }

Requirements: software
Requirements: hardware
Basic installation
Demo

Requirements: software

Requirements for Caffe and pycaffe (see: Caffe installation instructions)

Note: Caffe must be built with support for Python layers!

# In your Makefile.config, make sure to have this line uncommented
WITH_PYTHON_LAYER := 1
# Unrelatedly, it's also recommended that you use CUDNN
USE_CUDNN := 1

You can download my Makefile.config for reference. 2. Python packages you might not have: cython, python-opencv, easydict

Requirements: hardware

You will need a GPU to run the code (does not work in CPU-only mode). The present code is memory-efficient in the forward pass, but you may modify the .prototxt file to further reduce memory usage (please consult the paper, ubernet/test.prototxt and net.cpp to understand how the "deletetop", "deletebottom" variables are used)

Installation (sufficient for the demo)

Clone the UberNet repository

git clone https://github.com/jkokkin/UberNet.git

We'll call the directory that you cloned UberNet into UBERNET_ROOT
Build the Cython modules
```
cd $UBERNET_ROOT/lib
make
```

Build Caffe and pycaffe

cd $UBERNET_ROOT/caffe-fast-rcnn
# Now follow the Caffe installation instructions here:
#   http://caffe.berkeleyvision.org/installation.html

# If you're experienced with Caffe and have all of the requirements installed
# and your Makefile.config in place, then simply do:
make -j8 && make pycaffe

Download a trained UberNet model from here: https://www.dropbox.com/s/fbeg10aoicn4wc4/model.caffemodel?dl=0 and place it under $UBERNET_ROOT/model

Demo

To run the demo

cd $UBERNET_ROOT
./demo/demo_ubernet.py

takmin/UberNet