Validation Data, Loss and Overfitting

I see during training trainval.txt is used https://github.com/Megvii-BaseDetection/YOLOX/blob/6880e3999eb5cf83037e1818ee63d589384587bd/yolox/data/datasets/voc.py#L105
and after each epoch we evalute the model on test data instead of validation data https://github.com/Megvii-BaseDetection/YOLOX/blob/6880e3999eb5cf83037e1818ee63d589384587bd/exps/example/yolox_voc/yolox_voc_s.py#L104 I mean if i am training the model and I need to check if it is overfitting or not, how can I decide. May be I am wrong if this is not the way done in YOLOX. Actually I am doing some data sampling strategy but only on train data and I want to see if it is over-fitting the model or not.

Start with yolox/data/datasets/voc.py around line 105 and exps/example/yolox_voc/yolox_voc_s.py around line 104, then trace how trainval.txt and test data are selected during training and evaluation. Run the VOC training and evaluation flow to verify whether test data is used for per-epoch evaluation. Done means the intended validation behavior and its effect on checking overfitting are established.