1
00:00:00,053 --> 00:00:03,963
[MUSIC]

2
00:00:03,963 --> 00:00:08,103
We saw how deep features were great for
image classification.

3
00:00:08,103 --> 00:00:12,829
They allowed us to get quite high
accuracy, even with just a little bit of

4
00:00:12,829 --> 00:00:17,247
training data on a particular
classification task by first finding

5
00:00:17,247 --> 00:00:22,843
features from this deep learning model
that won the 2012 ImageNet competition.

6
00:00:22,843 --> 00:00:24,740
So, that was pretty exciting.

7
00:00:24,740 --> 00:00:27,710
But if you remember
the motivation at the beginning,

8
00:00:27,710 --> 00:00:30,985
kind of the sample use case at
the beginning of this module,

9
00:00:30,985 --> 00:00:34,890
we're talking about finding
images of products that I like.

10
00:00:34,890 --> 00:00:38,470
You know, I like shoes, I'm gonna find
shoes that's similar to my shoes.

11
00:00:38,470 --> 00:00:43,180
I want to find images that are similar
to in the product set I care about.

12
00:00:43,180 --> 00:00:45,520
So instead of doing an image
classification task,

13
00:00:45,520 --> 00:00:49,160
now we're gonna take those deep features
and do an image retrieval task.

14
00:00:49,160 --> 00:00:51,850
So let's go ahead and do that, okay?

15
00:00:51,850 --> 00:00:57,018
So we're going to name our notebook, so

16
00:00:57,018 --> 00:01:04,470
I'm gonna talk about Deep Features for
Image Retrieval.

17
00:01:04,470 --> 00:01:09,466
And I'm going to hide the header and
hide the toolbar,

18
00:01:09,466 --> 00:01:14,700
so we can go ahead and
get a little bit more space.

19
00:01:14,700 --> 00:01:20,760
And I'm gonna fire up graphlab create and

20
00:01:22,680 --> 00:01:28,861
I'm also going to load the CIFAR-10

21
00:01:28,861 --> 00:01:34,150
dataset, which is the same
dataset that we loaded

22
00:01:34,150 --> 00:01:39,440
earlier when we talked about deep
features for classification.

23
00:01:39,440 --> 00:01:44,142
So in this case I'm just going to
load the training set because we're

24
00:01:44,142 --> 00:01:47,400
just going to do retrieval
in the training set.

25
00:01:48,640 --> 00:01:50,390
So let's start from there.

26
00:01:50,390 --> 00:01:55,060
So I'm gonna call it image_train and

27
00:01:55,060 --> 00:02:00,130
I'm going to use graphlab.Sframe and

28
00:02:00,130 --> 00:02:07,540
then as input I'm going to
give it the image_train_data/.

29
00:02:07,540 --> 00:02:14,576
And just as a little reminder,
if I look at this image_train data set,

30
00:02:14,576 --> 00:02:19,265
.head, in this s frame,
we have the image id,

31
00:02:19,265 --> 00:02:25,015
the actual image,
the label which we're not gonna use for

32
00:02:25,015 --> 00:02:30,670
retrieval, we're not gonna
use that label at all.

33
00:02:30,670 --> 00:02:34,030
The image array of the pixels,
which we're also not gonna use, but

34
00:02:34,030 --> 00:02:36,590
it already has the deep
features precomputed.

35
00:02:36,590 --> 00:02:41,520
In the notebook on deep features for
classification,

36
00:02:41,520 --> 00:02:44,950
we looked at those lines that you need
to compute that column, deep features.

37
00:02:44,950 --> 00:02:49,520
Here, I'm just gonna use
those deep features.

38
00:02:49,520 --> 00:02:54,029
[MUSIC]