Benchmarking Pastec with the Stanford Mobile Visual Search Data Set

On the personal page of David M. Chen from the Stanford university, a very interesting data-set is available. The Stanford Mobile Visual Search Data Set consists of a collection of reference images (book covers, business cards, CD covers, DVD covers, landmarks, paintings and printed documents) and query images taken with a mobile device or a camera with various points of view and lighting conditions.

Achieving good results on this dataset is quite challenging as the query images have intentionally been taken in conditions far from being optimal. I thought it would be a good occasion to benchmark Pastec and show an example usage of the Python 3 PasteLib.
The following code indexes all the DVD cover reference images and tests that the query images match.

#!/usr/bin/python3

import PastecLib

DATASET_PATH = "/path/to/dvd_covers"

PATH_REFERENCES = DATASET_PATH + "/Reference/"

PATH_QUERIES = DATASET_PATH + "/Canon/"
#PATH_QUERIES = DATASET_PATH + "/Droid/"
#PATH_QUERIES = DATASET_PATH + "/E63/"
#PATH_QUERIES = DATASET_PATH + "/Palm/"

c = PastecLib.PastecConnection()

# Index the reference files.
for i in range(1, 101):
    c.indexImageFile(i, PATH_REFERENCES + "%03d.jpg" %i)
    
goodMatches = 0    

# Query the index.
for i in range(1, 101):
    res = c.imageQueryFile(PATH_QUERIES + "%03d.jpg" %i)
    if res != [] and res[0] == i:
        goodMatches += 1
    else:
        print("Image n°%d not matched." % i)

print("Number of good matches: %d" % goodMatches)

Quite simple, no? ;-)

Displee, the Pastec demonstrator!

Until now Pastec, our open source image recognition index and search engine for mobile apps, lacked a real demonstrator. This gap is now filled by Displee! Displee is a tiny Android app that allows to search pages in the English Wikipedia by taking pictures. It is a kind of open source Google Goggles for images from en.wikipedia.org.

Get it on Google Play

The Pastec index of Displee is hosted on a single server. It currently contains about 440 000 images. They may not be the most relevant ones but this is a start. The good news is that Displee is also released in open source (MIT licence).

I think Displee is a nice use case that shows well the possibilities offered by Pastec.
I hope you will enjoy it!

Displee's screenshots

Initial release of Pastec

We are very proud today to release the first version of Pastec, your free image recognition platform.

We hope that you will like this website and that you will be able to find all the information you need.

For any request, do not hesitate to mail us at contact@visualink.io.