Hi, this post might include very basic and silly questions since I am from the art field. I'm looking for solutions for my socially-engaged art project in Helsinki and after done tones of research, I realized that prodigy might be the killer app for me.
The basic scenario is to invite the public audience to join the labeling activity which target to my dataset of AI-generated abstract images(training with amount of sensitive social/political topics hashtags on the social media ). a sample shows below:
It's kind of conceptual interactive art rather than training a valid model in the datascience field, so the definition needs to be freely entered by the audience after drawing regions of box/polygon rather than predefined;
second, like the demo page shows, people could join the annotation activity without extra efforts (the workflow of setup and invite team members)by visiting a public URL with their smartphones. Is this easy to achieve through personal plan?(unlimited annotators means I don't need to implement prodigy teams, right?)
And as far as I know, since prodigy doesn't hold any data from us, which is very different from most of the other online web-based data labeling platforms, the next question would be, It seems that I need to purchase a plan from some web design platforms or just some DB server service in order to deploy/ publish my customized interface in the Python environment?
Appreciate for any suggestions or just a link of a tutorial video for a dummie would be welcome!
Hi! This is definitely and interesting concept and I'm glad to hear Prodigy looks interesting to you
To be completely honest, I'm not sure Prodigy will be the right fit for what you're trying to do. Prodigy has been very specifically designed as a developer tool for creating training data for machine learning models. This means that the way the tool works and the assumptions it makes are all focused on making the process of creating datasets and running machine learning experiments as robust, consistent and efficient as possible.
Prodigy also comes with a lot of built-in best practices for machine learning, which is really useful for data scientists doing applied work – but which might actually be kinda counter-productive for the sort of artistic use case you're describing. For example, if you're collecting data to train a model, you typically wouldn't want the annotator to just freely enter labels – usually, the best way to get accurate results is to design the label scheme upfront and with the model architecture in mind. You'd also want things to be as structured as possible, and automate the work wherever you can – which is again very different from collaborative art.
This is also a use case Prodigy isn't really designed for and that the license wouldn't fully cover. You can definitely host the app on your server (e.g. any cloud hosting provider) and send the link to your annotators, and there's technically no limit on how many people can annotate for you. But you couldn't just post a link on the internet and ask anyone to join.
This is another example of a workflow that makes a lot of sense for an art project, but less so for applied machine learning: for many tasks, it's more effective to do the annotation with a small number of annotators and/or subject matter experts in-house, rather than asking anyone on the internet to create arbitrary data.
ThanX form the bottom of my heart for the quick response. It's really a detailed explanation and clarification from a developer's view and experience! I will check my plan and looking forward to seeing your next giant leap to success anyway~
Hi, @ines
Based on your feedback, I've tried to redefine the workflow of my project by
People join the collaborative activity by login the page of web service with passwords (not by sending application to join or approved by the administrator)
Setting Few categories of labels for choosing (not by object names)
Users add annotations (box/polygan/brush) and add some compulsory captions to describe their thoughts
Does it sound more reasonable for proceed with the personal licence? ThX for your time again~
These are definitely things you can do with Prodigy, and if you select the people who get to annotate (and don't just post the password on the internet), there shouldn't be a problem with the license. I obviously can't comment on how useful it will actually be for your use case, since it's quite different from the machine learning projects we usually think about
Since it's a developer tool, setting up and hosting Prodigy does require some general knowledge of Python, the command line and probably servers (if you want to host it on a server). So if you don't have experience with that, maybe you can ask a friend or collague to help with this. It's all very standard technologies, though, and none of it is really specific to Prodigy.
