biolab / biolab/orange3-imageanalytics

Default embedder sends data externally without clear indication

Open
#254 0 comments 1 reaction 0 assignees View on GitHub
Dominant language
Python
Stars
35
Forks
42
PR merge metrics
No merged PRs in 30d

Description

I've been investigating Orange for Image Analytics as a potential tool for teaching basics. However I noticed something that extremely concerned me with this plugin.

The default embedder (Inception v3) sends images to an external server, but this isn't obvious in the UI or in a number of youtube tutorials - I only discovered it after digging deeper. I have since seen that this is mentioned on FAQs and on the widget website, but it is not very clearly stated. A user trying the software without reading documentation or digging into settings won't realise their data is being sent externally.

There are many closed datasets that should not be sent to any third party for any means. Some of the ones used for teaching definitely shouldn't be. Thankfully I can guide my students to not use online ones, but it's not intuitive.

I see PR (https://github.com/biolab/orange3-imageanalytics/pull/252) is aiming to add more local models which will help, but my concern still stands regarding defaults.

Here are my general thoughts/suggestions, although it is your project so feel free to take or leave any/all.

- Label online embedders explicitly e.g. `Inception v3 (Online)`.
- Add more visible note in the widget pages that data may be sent externally depending on embeddings and that the user should check to ensure they are allowed to
- Have an offline/local model as the default

It's great that you do offer online embeddings for those with limited hardware, I just think it would be better to be more explicitly mentioned and opt-in. Great work otherwise.

Contributor guide

No contributing guide indexed for this repository

Research direction

Start by reviewing the default embedder and the widget pages, then read PR #252 for related local-model work. Clarify which combination of explicit online labeling, visible data-transfer guidance, and an offline default is required, and define done with maintainer agreement on the intended default behavior.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
documentation, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.