Upload
others
View
12
Download
0
Embed Size (px)
Citation preview
The VGG Image Classification (VIC) Engine
● Performs a visual search over a dataset of images with categories ● Automatically computes an image classifier to find images matching your query
within the dataset ● Input can be a text string or an image ● It can be used to make your images searchable
How to make searchable images from IIIF repositories
1. Start VIC
2. Remove previous images and metadata (OPTIONAL)
3. Login to the administrative tools
4. Stop the backend engine
5. Remove previous backend data (OPTIONAL)
6. Start the IIIF image downloading process
7. Input data download and check
8. Wait for the image processing pipeline to finish
9. Start the backend engine and reset the metadata service
10. Search your images !
1. Start VIC
Start VIC using Kitematic as described in
Mac: http://www.robots.ox.ac.uk/~vgg/software/vic/docker_deployment_mac.html
Windows: http://www.robots.ox.ac.uk/~vgg/software/vic/docker_deployment_win.html
Linux: Start VIC using the command-line or the helper scripts, as described in
http://www.robots.ox.ac.uk/~vgg/software/vic/docker_deployment_linux.html
2. Remove previous images and metadata
If you are ADDING ADDITIONAL IMAGES to your dataset, skip this step. Otherwise,
remove any previous files in [MY_FOLDER]/vgg/mydata/images/mydataset
(WARNING: Remove the contents of the folder but not the folder itself)
Also remove the contents (not the folder itself) of
[MY_FOLDER]/vgg/vic/frontend_data/searchdata
Please take into account that you are about to download
images from an IIIF repository and they can be quite large.
Therefore, please make sure that your docker service has
enough memory allocated (at least 4 GB) before attempting
the data ingestion, see
Win:https://docs.docker.com/docker-for-windows/#advanced
Mac: https://docs.docker.com/docker-for-mac/#advanced
2. Remove previous images and metadata
If you are ADDING ADDITIONAL IMAGES to your dataset, skip this step.
However, if you want to clear completely your previous metadata, remove the
[MY_FOLDER]/vgg/mydata/metadata/indexdir folder (if present) as well
as any previous file in [MY_FOLDER]/vgg/mydata/metadata/mydataset
(WARNING: in this case do not remove the folder itself).
Remove this folder completely (if present)
Remove just this file and other files INSIDE the folder,
but not the folder itself
3. Login
Use the “Sign In” link in the top-left corner of the VIC home page to go to the login
page. If you are already signed in, use the “Admin Tools” link and go to Step 4.
Use the credentials
user: admin
passwd: vggadmin
4. Stop the image classification engine
Go to the Manage Backend Service tab and Stop the backend service
5. Remove previous backend data
If you removed files in step #2, you should remove the existent backend data. Go
to the Manage Backend Data, and under Clear Backend Data make sure
the Search Engine is set to Category and then press the Clear button. This is
NOT necessary if you are just adding new images
6. Start the processing of your images
Go to Manage Backend Data. Under Ingest New Data make sure the Search
Engine is set to Category and Input Type is Images
Then, specify an IIIF Manifest
source and press Ingest
In this case, select a file in your
local storage
In this case, type the URL
pointing to an IIIF repository
Only one IIIF source can be
used at a time
7. Input data download and check
VIC will download the images from the IIIF source that you specified and will check if
they are valid. Please be aware that if the IIIF source specifies a large number of files
or there are any delays in downloading the files from the Internet, this process might
take a long time.
8. Wait for the image processing pipeline to finish
Wait until your images are processed (100 images in the example below)
9. Start the backend engine and reset the metadata
Go to the Manage Backend Service tab and Start the backend service
9. Start the backend engine and reset the metadata
Go to the Manage Backend Service tab and Reset the metadata service
10. Go back to the search page
Click on the Home link to go back to the main page and start a query !
Remember to Stop VIC
Stop VIC using Kitematic as described in
Mac: http://www.robots.ox.ac.uk/~vgg/software/vic/docker_deployment_mac.html
Windows: http://www.robots.ox.ac.uk/~vgg/software/vic/docker_deployment_win.html
For Linux: Stop VIC using the command-line or the helper scripts, as described in
http://www.robots.ox.ac.uk/~vgg/software/vic/docker_deployment_linux.html
Data Ingestion Tips
1. Images in IIIF repositories can be quite large, so in order to speed up the data ingestion as well
as the image queries, the downloaded images are scaled down to a maximum of 500 pixels in
width.
2. If you just want to ADD A FEW IMAGES to your dataset, you can choose to “Select files from a
local storage” in step #6 and press the “Choose Files” button. Then select manually the files you
want to add. Note that you can only select files, not folders. The new files will be automatically
copied to [MY_FOLDER]/vgg/mydata/images/mydataset.
3. If you want to add a large number of local images, and you choose to ingest the complete
images folder in step #6, the input data check in step #7 might take a long time because the
COMPLETE images folder is scanned to acquire the paths to the files. To avoid this, create a
text file (with extension “.txt”) with the paths to your image files, relative to the folder
[MY_FOLDER]/vgg/mydata/images/mydataset. After that choose to “Select files from a
local storage” and then press the “Choose Files” button shown in step #6 and specify the text
file you created. This will avoid the folder scanning and the data ingestion will start immediately.