Does validation data influence the results of YOLO V2 and faster R-CNN?

I am training YOLO V2 and faster R-CNN using my own dataset. I found that YOLO V2 gets a much better result when validation data are not used, and faster R-CNN gets a little better result when validation data are not used. The dataset has been devided into train dataset, validation dataset and test dataset ealier, so no matter the validation data are used or not, training data and test data is all the same. So I think that if it is the validation data influence the results of training a detector? and how ?

Answer 1

[Shashank Gupta]

Hi,

Validation data doesn't influence the results untill and unless the data is used during the training. validation data is handled separately and there is possibly no overlap in training. the results might be varying because of randomness in the model. Try running it multiple time to see. Nevertheless, I will investigate it on my side too, But do check your experiments again, and reach out if you see same behaviour.

[tara saber]

how to use validation data in Yolo v3 ?