Skip to content

Commit

Permalink
Merge branch 'master' of https://github.com/yaolubrain/ULNNO
Browse files Browse the repository at this point in the history
  • Loading branch information
yaolubrain committed Oct 7, 2015
2 parents 1b76aaf + f2d4e1c commit c134943
Showing 1 changed file with 7 additions and 5 deletions.
12 changes: 7 additions & 5 deletions README.md
Original file line number Diff line number Diff line change
@@ -5,30 +5,32 @@ This repo contains the experiment code in paper

The paper presents a new zero-shot learning method, which achieves the state-of-the-art results in ImageNet fall2011.

The CNN model is GoogeLeNet with [Caffe] (http://caffe.berkeleyvision.org/) implementation. The image format convertor (image2hdf5) is from [Toronto Deep Learning](https://github.com/TorontoDeepLearning/convnet).

## Instructions
### download the following files from http://image-net.org/
- ILSVRC2012_img_train.tar (138G)
- ILSVRC2012_img_val.tar (6.3G)
- fall11_whole.tar (1.2T)

### prepare the images intro HDF5 files, use
### prepare the images into HDF5 format with
- uncompress.sh
- correct_format.sh
- image2hdf5.sh

### compute the CNN outputs of GoogLeNet of the images, use
### compute the CNN outputs of GoogLeNet of the images with
- caffe_outputs.py

### compute PCA and ICA on the CNN outputs, use
### compute PCA and ICA on the CNN outputs with
- cov.py
- whitening.py
- ica.py

### compute the MDS features of WordNet graph, use
### compute the MDS features of WordNet graph with
- similarity_mat.py
- mds_distance_mat.m

### run zero-shot learning experiments, use
### run zero-shot learning experiments with
- imagenet_1k_21k_idx.py
- imagenet_zero_shot_unseen_wnids.py
- make_zero_shot_mat.m

0 comments on commit c134943

Please sign in to comment.