imagenet

A python module containing both a full imagenet dataset object conforming to skdata standards, and various related subsets.

This folder will be created when using datasets: ~/.skdata/imagenet/images

Which is where the images will may be cached locally (not default, see dataset.get_images for documentation)

You will need a user account on our database (or a similarly configured one). Email [email protected].

To install:

$ pip install git+http://github.com/dicarlolab/imagenet.git#egg=imagenet

or if you don't have root access

$ pip install --user -e git+http://github.com/dicarlolab/imagenet.git#egg=imagenet

you have to install the requirements from the requirements file as well

pip install -r requirements.txt

tunnel to the database

ssh -f -N -L 27017:localhost:27017 [email protected]

you must also use the nltk package to download wordnet

nltk.download('wordnet')

Some examples:

Import the dataset and call its constructor

import imagenet.dldatasets as d
dataset = d.Challenge_Synsets_20_Pixel_Hard()

The dataset has a meta tabular array object

meta = dataset.meta

And a dictionary containing a dictionary of information about each synset, each of which is represented by a wordnet id

synset_meta = dataset.synset_meta
list_of_wordnet_ids = synset_meta.keys()
info_about_first_synset = synset_meta[list_of_wordnet_ids[0]].keys()

get_images() can use the dataset.default_preproc spec

images = dataset.get_images(dataset.default_preproc)