Is there any way to include negative images to enhance the accuracy

experiencor / keras-yolo2

Easy training on custom dataset. Various backends (MobileNet and SqueezeNet) supported. A YOLO demo to detect raccoon run entirely in brower is accessible at https://git.io/vF7vI (not on Windows).

MIT License

1.73k stars 784 forks source link

Is there any way to include negative images to enhance the accuracy #401

Open ghost opened 5 years ago

ghost commented 5 years ago

Hello Sir, Thanks so much for your effort. please is there any way to add negative samples which do not contain the desired object to enhance the prediction. Is that possible ?. Is that will enhance the accuracy.

MBoaretto25 commented 5 years ago

@hkhaledmohamed Yes, you can do that as is suggested in here for the racoon problem. The model must learn what is and what is not the object that it will be set to predict.