I am trying to retrain a neural network using transfer learning that can classify whether an image has a certain object, say, a car. My positive sample dataset is quite small, only 2500~ images. It works really well with "regular" binary classification (2500 images of cars/2500 images of flowers and it has to differentiate between those two), but the problem is that I am not sure how to make it classify for all types of images, or how to make it 2500 images of cars/2500 random images and it would have to classify whether Yes - the image is a car or No - it is not a car. What would those random images be? Thanks.
$\begingroup$
$\endgroup$
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
1
Have you tried creating a dataset with 2500 images of cars and ~2500 images of random stuff and trained it first? if so you should try that and see how it works, if it doesnt work you might have to add more images in the "distractor" class. Provide more information about what you find after you do this experiment and may be we can help further?
-
$\begingroup$ I did! Sorry for not mentioning that. I downloaded batch images of nature, dogs, food, etc. I think that might have been the problem? The images that I downloaded could have been too similar to each other so it would have ended up being a bunch of batches of similar images. It failed, miserably. It thought that a hotdog was a car. Basically ended up having a very high bias towards the positive examples. That's where I got lost and started doubting if it's even possible to do with such a small dataset of positive examples. $\endgroup$– user90289Feb 19, 2020 at 13:04
$\begingroup$
$\endgroup$
It depends on the goal of the project. What is a "not car"? A motorcycle? A truck? A van? A billboard with a picture of a car?
Within the project's goals, the "random" images should be as close to a car as possible and as diverse as possible.
answered Sep 6, 2022 at 16:18