I have data with 1080 width and 1920 height values. I want to train a model according to these input shapes. Because I will use it on android. I don't want to size the photo as 640 640, which is the default size of the path for detection in small objects. How can I do this? So how can I train the model so that the input tensor is 1080 1920?
I use yolov7 with this code: !python train.py --batch 8 --img 1920 --epochs 200--cfg /content/drive/MyDrive/AI/yolov7/cfg/training/yolov7.yaml --hyp /content/drive/MyDrive/AI/yolov7/data/hyp.scratch.p5.yaml --data /content/drive/MyDrive/AI/ndatalast/a/data.yaml --weights 'yolov7_training.pt'
When I train it this way, the model success is very bad.
