Using same data for Training and Testing

Hi There, I wonder what would happen if I use the same Image I use for training process in the testing ? Is this some kind of bad practice ? Since the dataset is very specific.

Just to visualize it better, if I train a network to recognize my self I will put as much of info for every angle possible for it to train that it’s me. And now I want to make sure that in every angle that I taken the Network can recognize me hence why I use the same Image for testing.

Is my question make any sense ? Since I want to know if it kind of bad idea that will effect the training accuracy. FYI I’m usally just split the image but since the testing image is kind of not being trained (too far different angle) so the result it not that good for that specific angle but good for the other angles

After reading to the documentations provide by the amazing team at I came to conclusion that using the different train and test set will help in the evaluation process for the network to better generalize the object.

It will really come in handy when you want to know whether you should continue training or not. There is an excellent documentation about Early Stopping here Early Stopping - Deeplearning4j that will help you better understanding to optimize the network you’re training.

I’ll leave the Topic here in the forum so if anyone else wonder the same question that would not need to ask anymore and anyone with better knowledge than I do can come and give us more information about the topic.

I’ll leave everything to the forum admin to decide whether to close this topic or not.
Thank You, Have a wonderful day and keep on learning :grin: