|
1
|
|
|
# -*- coding: utf-8 -*- |
|
2
|
|
|
from fuel.datasets import H5PYDataset |
|
3
|
|
|
from fuel.transformers.defaults import rgb_images_from_encoded_bytes |
|
4
|
|
|
from fuel.utils import find_in_data_path |
|
5
|
|
|
|
|
6
|
|
|
|
|
7
|
|
|
class ILSVRC2010(H5PYDataset): |
|
8
|
|
|
u"""The ILSVRC2010 Dataset. |
|
9
|
|
|
|
|
10
|
|
|
The ImageNet Large-Scale Visual Recognition Challenge [ILSVRC] |
|
11
|
|
|
is an annual computer vision competition testing object classification |
|
12
|
|
|
and detection at large-scale. This is a wrapper around the data for |
|
13
|
|
|
the 2010 competition, which is (as of 2015) the only year for which |
|
14
|
|
|
test data groundtruth is available. |
|
15
|
|
|
|
|
16
|
|
|
Note that the download site for the images is not publicly |
|
17
|
|
|
accessible. To download the images, you may sign up for an account |
|
18
|
|
|
at [SIGNUP]. |
|
19
|
|
|
|
|
20
|
|
|
.. [ILSVRC] Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, |
|
21
|
|
|
Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya |
|
22
|
|
|
Khosla, Michael Bernstein, Alexander C. Berg and Li Fei-Fei. |
|
23
|
|
|
*ImageNet Large Scale Visual Recognition Challenge*. IJCV, 2015. |
|
24
|
|
|
|
|
25
|
|
|
.. [SIGNUP] http://www.image-net.org/signup |
|
26
|
|
|
|
|
27
|
|
|
Parameters |
|
28
|
|
|
---------- |
|
29
|
|
|
which_sets : tuple of str |
|
30
|
|
|
Which split to load. Valid values are 'train' (1.2M examples) |
|
31
|
|
|
'valid' (150,000 examples), and 'test' (50,000 examples). |
|
32
|
|
|
|
|
33
|
|
|
""" |
|
34
|
|
|
filename = 'ilsvrc2010.hdf5' |
|
35
|
|
|
default_transformers = rgb_images_from_encoded_bytes(('encoded_images',)) |
|
36
|
|
|
|
|
37
|
|
|
def __init__(self, which_sets, **kwargs): |
|
38
|
|
|
kwargs.setdefault('load_in_memory', False) |
|
39
|
|
|
super(ILSVRC2010, self).__init__( |
|
40
|
|
|
file_or_path=find_in_data_path(self.filename), |
|
41
|
|
|
which_sets=which_sets, **kwargs) |
|
42
|
|
|
|
mila-udem /
fuel
