PDF Archive

Easily share your PDF documents with your contacts, on the Web and Social Networks.

Share a file Manage my documents Convert Recover PDF Search Help Contact



Copy of Info259 Poster .pdf


Original filename: Copy of Info259 Poster.pdf

This PDF 1.4 document has been generated by Google / , and has been sent on pdf-archive.com on 25/12/2017 at 11:19, from IP address 122.163.x.x. The current document download page has been viewed 129 times.
File size: 346 KB (1 page).
Privacy: public file




Download original PDF file









Document preview


DeepSeek:

Content Based Image Retrieval
Tanya Piplani (tanyapiplani@berkeley.edu)

Introduction

Models
Caption Based Retrieval

Quantitative Metrics
Caption Generation Model
Model

BLEU-1

METEOR

ROUGE-L

CIDEr-D

Our

0.828

0.280

0.603

0.692

SOTA

0.953

0.375

0.734

1.270

Pr@5

Time

Image Retrieval Models
Model

DeepSeek a natural language processing based deep
learning model that allows users to enter a description of
the kind of images that they want to search, and in
response the system retrieves all the images that
semantically and contextually relate to the query.

Embedding Based Retrieval

Pr@1

Pr@3

Caption
Based

0.729

0.845

0.905

3.89 sec

Embedding
Based

0.683

0.857

0.912

4.22 sec

Data
Caption Generation Model
MS COCO 2015 dataset was used for training the
caption generation model. 80k training images, 20k
validation and 20k test images.

Image Retrieval Models
For training the Embedding based retrieval model, the
generated captions on a subset of the train set of MS
COCO 2015 along with its images were used (10k
images + captions). The test set consisted of a subset
of the test set of MS COCO (10k images + captions).

Analysis & Conclusion
We see that both the caption based retrieval system
and the embedding based retrieval system do a
good job at content based image retrieval. The
embedding system while slow, due to the need of
calculation of embeddings at query time, is slightly
more accurate when precision@5 is considered.
More training and GPU optimization could make it
faster and more accurate.


Document preview Copy of Info259 Poster.pdf - page 1/1

Related documents


copy of info259 poster
22i20 ijaet0520903 v7 iss2 477 487
ijetr011720
how to create great ebooks
8i20 ijaet0319427 v7 iss2 359 371
viooz


Related keywords