# Building Image Dataset In a Studio

I'm currently working in a problem of Object Detection, more specifically we want to count and differentiate similar species of moths.

We are already testing some detection algorithms:

For the datasets, first we'll scrape the web for those images to have a starting model.

Then we'll get plenty of pictures from our devices that will look like this:

Unfortunately it will take 6 months to get those pictures.

Currently we have some samples of moths, and we have the resources to take pictures of them in a studio. We want to simulate the field conditions as well as taking close ups of the moths in different angles and lighting conditions. This experiment would also allow for us to get the accuracy of classification from an expert entomologist.

Is the effort of taking studio photos worth it?

Would these pictures significantly increase the models accuracy?

Is there any opportunity for publication?

Is there any paper that uses studio photos as well as field photos for object detection?

Also, there's plenty of opportunities for publication here in the company, so they would go the extra mile if we publish something, or if these pictures would improve the final accuracy of the models.

• I gave you a answer, but be aware this kind of question is primaly opinion based and can't have a definitive answer. Also, even tho you plan to use Convolutional Neural Networks or any Deep Learning Tools, your question does not have a mention of these methods, that said I think it is not correct of you to use those flags. – Pedro Henrique Monforte Mar 28 '19 at 2:31
• Thanks for the answer and the feedback. I will be editing the question to include the methods we'll try. – Luiz Amaral Mar 28 '19 at 12:21

## Disclaimer:

This is a question that is probably going to be flagged since it is too broad and answers will be mainly based on opinion.

## Is this effort worth it?

That is subjective, what commercial/social use does this have? Is this kind of detection really relevant? What can you do with those detections and classifications? That is something only an entomologist could answer properly, you will get better answers for this one with them.

If you can find this kind of motivations for your work, you probably get a good dataset with good use. But if you don't even know what kind of applications you're aiming to you should start doing more research on motivation.

## Would these pictures significantly increase the model's accuracy?

It is highly probable that it will, but you can't be sure until you try. Also, train your models with recently acquired images during the acquisition process to make sure that those images are having an impact on your models.

## Is there any opportunity for publication?

Surely there is, if it will be cited multiple times or if the paper will be well received by Machine Learning community? Probably. If not, that probably would get you a top publication on entomology.

## Is there any paper that uses studio photos as well as field photos for object detection?

Well, there are many. These controlled datasets are pretty easy to come up with and have been used extensively see BioID for example, until today many publications use it for benchmarking on Eye Pupil Detection and Facial Landmark Detection. In the wild datasets (usually collected from the web) are usually more challenging and also more fitted to real-world application, but it is not always possible to acquire them easily.

Last year CEFET-RJ and UFRJ published papers in IEEE with created datasets for Aedes Aegypti breeding sites detection. This dataset was created using drones and manually annotated by CEFET-RJ and UFRJ undergrad students.