[
https://issues.apache.org/jira/browse/TIKA-2398?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16068484#comment-16068484
]
Thamme Gowda commented on TIKA-2398:
------------------------------------
[~ThejanWijesinghe] Looks good. This will be the next milestone for GSoC.
> Unifying Object Recognition REST services
> -----------------------------------------
>
> Key: TIKA-2398
> URL: https://issues.apache.org/jira/browse/TIKA-2398
> Project: Tika
> Issue Type: Improvement
> Components: parser
> Affects Versions: 1.16
> Reporter: Thejan Wijesinghe
> Labels: REST_API, deeplearning, gsoc2017, machine_learning
>
> h1. Background
> Tika already has a [Object Recognition
> Parser|https://wiki.apache.org/tika/TikaAndVision], a [Video Labeling
> Parser|https://wiki.apache.org/tika/TikaAndVisionVideo] and an ongoing
> [PR|https://github.com/apache/tika/pull/180] for an Image Captioning Parser.
> All of these parsers are based on REST services, but currently there's no
> convenient way users can deploy all these REST services, There can be use
> cases such as user needs to use the Object Recognition Parser and the Image
> Captioning parser together. The entire implementation will be based on
> docker. Once implemented this, user will not have to build every single
> docker for object recognition, video labeling, image captioning etc. etc.
> User will only need to build the docker container which has this unified REST
> server and web gui. This docker container will build other docker containers
> for image captioning, object recognition etc. etc and host them as REST
> services whenever user needs them. For an example, if a user needs to use
> object recognition parser for the first time, he will have to only run this
> docker container(unified REST server with web gui), then user can activate
> object recognition service from the web gui. First time activating that
> service, it will automatically create the docker container for object
> recognition, then will make object recognition service available to the user.
>
> h1. Objectives
> # Automating the docker container building process for the user
> # Creating a convenient platform for the user where he can start/terminate
> REST services, see statistics of the model usage through a web gui
> # Making the current REST services more stable using a reverse proxy server
> h1. Benefits
> # Convenience
> # Easy to implement a new parser with deep learning capabilities by finding
> an already trained NN model with any framework. i.e.framework doesn’t matter.
> We don't need to stick to Keras, Tensorflow or DL4J. If we can host the
> service as a web app. This platform can support it.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
