Dennis Lin
Reputation: 65
Can CNN recognition answer "Not any one of these classes"
I use CNN to do image recognition for twenty classes of images. I also want CNN to tell me when the image is not belong to any one of these twenty classes. Softmax can only give me the answer of one of these twenty classes. Is there any way to solve the problem?
Upvotes: 0
Views: 113
Answers (1)
theastronomist
Reputation: 1056
You should create a "negative class" of random objects that are not your 20 classes. Training on negative data should increase accuracy and generalization because it can help the network figure out the object from a background
Upvotes: 1
Related Questions
- Why do sometimes CNN models predict just one class out of all others?
- CNN is not able to classify images
- Identifying images not belonging to the classes the CNN is trained with
- CNN not predicting class of image
- Can a Keras CNN predict multiple classes?
- CNN Model With Low Accuracy
- CNN classifier on dataset with one random class
- What's keeping this simple CNN from classifying whether an image contains a cat or not?
- How to categorize image as a 'not categorized image' by using convolutional nets
- Convolutional neural network doesn't classify test set keras