Hello everyone,

I am a pure beginner with deep learning.

I have used NVIDIA DIGITS to train a neural network against my own image dataset. I have managed to obtain very strong accuracy, the system always detects the right objects within my own image dataset. However when I provide the system with an image which doesn’t contain any objects from my image dataset, it still gives me high accuracy percentage. Is there any way of getting the accuracy percentage lower given I provide the system with an image which doesn’t contain an object within my image dataset.

I’d appreciate any sort of help and apologies if I was unclear.

Is this still an issue? If so I have a few questions.

I don’t understand what you are testing here. When you say you “obtain very strong accuracy” are you saying the accuracy of the your validation set is high?

You then are using a single image to do inference on with your trained model. When you perform inference with an image containing an object not in your original dataset, you get a high accuracy to one of labels in your original dataset.

Are you training with a classification or object detection network? Which one?