How to detect specific object effectively?

#1

I am working on a project to detect Caracal. It is an endangered animal also called wild cat. The issue is that my model accuracy is very low. I have tried different models such as such as FOMO and mobilnet SSD. By the way, I have some questions.

Do I have to create different object classes or just one class is enough?
How many images are required for good (85+) accuracy?

#2

Hello @ujjwalrathod007 first of all welcome to the Edge Impulse community!

Your project sounds amazing! Tell us more about your project:

  • where are you planning to deploy the cameras?
  • what are the lighting conditions?
  • are you getting low accuracy testing the model? are you using the same cameras for training and for inference?
  • how many images do you have? 1 object should be ok!

Please share more details so we will be able to help you more.

In the meantime, let me share as well the Image classification docs in case it gives you some ideas

#3

Hi,
I am building the prototype for now and based on success maybe I can decide.
But this is always in forest with low/no lights during night time.
I am not using the same camera for collecting the data. I collect data from online such as dataset and even google because there is not much data available for this specific animal.
At the moment I have 50 images for training and testing. All having at least one Caracal image. So I have only one label.

I am going to use Raspberry Pi for this model deployment.

Thanks for sharing the docs. I will have a look…