We had this here: Is there any pretrained model for emotion detection?

enter image description here


1 Answer 1


I don't think the problem here is specific to open source authors. In fact I think that is a dangerous starting point: When it comes to free software resources, we will get a split opinion base - engineers that create free software are volunteers creating value for all in the real world, much like site contributors here. In many ways they are heroes that should be celebrated.

The trouble with focusing on the value equation is that it puts the site voters into position of voting based on judging the worthiness of the product. I would fully expect a question placed by a Google employee about pre-trained image classifiers and answered with links to inception networks would get shut down as spam. So what is the difference?

The linked question is the example of what can go wrong with allowing resource request questions on a Stack Exchange site. Most new sites struggle with what to allow when someone has a problem to solve but all they want is a link to something off-site that solves the problem directly. It is very useful to have links to the site subject's introductory material. However, such questions can quickly bring in contributions to the site that have agendas to promote some product or idea. Even when the product is free, the promotion typically has an agenda to increase reputation of the contributor off site - converting Stack Exchange answers into incoming links to the product.

The nature of the question should set the scope for an acceptable answer - if the most appropriate answer is just a list of properties of the project plus a link and disclaimer, then the problem is with the question.

We should start getting stricter about resource requests on AI Stack Exchange:

  • Vote to close questions that ask for links to completed AI services, software products or projects, in order to just use them (as opposed to understand how they work)

  • Downvote "Gimme an AI that does X" questions.

I think we can still accept questions about papers on subjects as there is no real history of academic paper writers self-promoting via Stack Exchange. But other resource requests need to be accepted more cautiously.

The OP of the question should look for existing questions about emotion classification in video, and answer accordingly. That would be a valuable contribution. A basic but good question about emotion recognition on AI could be asking whether neural networks are the only high-performing model, or whether NN models represent anything interpretable, whether they can be reversed to generate images that show "archetype" emotional faces, what the loss function should be to discover emotional responses instead of classifying them using supervised learning etc. All those questions would require more than just a link to a project - a brief link (with the disclaimer, but without listing traits unrelated to the question) would be appropriate if the project could be used as an example.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .