I’m just getting started with this and would like to try detecting if an image from my bird feeder has a bird in it. I’ve done some training and it seems to work okay, but when I process an image with the model it never detects a bird. I’m wondering if the image files are just not going to work, or maybe they will only work with a huge amount of training. See example image:
Hi @strlng ,
The bird in the feeder closely resembles the background which might be a little bit difficult for it to be recognized. How many training images do you have? maybe you could try adding some more images and try having a diverse set of the birds in your area. You can then retrain your object detection model and see if the performance improves.
Regards,
Clinton
Thanks. I was thinking it would work well if I fed it lots of images of just the bird feeder, since many photos are empty, and many of birds at the feeder. It would be better to feed images of just birds and then it would hopefully learn to recognize them in my images?
Hi @strlng,
Images of birds in the feeder would work well in this case. Maybe could you share your project ID to have a peek of some the images you currently have?
Regards,
Clinton
ID is 90280. I removed all my images yesterday and have started to re-upload some after cropping them. I will add another batch today.
Training the model on a cropped image seems to have solved the problem. It is performing wonderfully now.
2 Likes
Hi @strlng,
Perfect! To even make it easier detecting birds in your feeder, you could as well try using object detection and compare the performance with the classification model you just built. You can follow these tutorials: https://docs.edgeimpulse.com/docs/object-detection or Counting objects using FOMO . Let me know if you come across any challenges.
Regards,
Clinton