EfficientDet GPU memory consumption

[Ronald-Kray]

I'm working with EfficientDet, Yolov5 on my custom dataset(GTX 2070 Ti 11G)
I don't know why EfficientDet consumes so much GPU memory for training.
There are some reasons for this as follows, but I don't fully understand.
Could you help me with these issues?

EfficientDets use a lot of GPU memory for a few reasons
For CNNs, memory usage is mostly dominated by activations rather than parameters.
Large input resolution: because resolution is one of the scaling dimension, our resolution tends to be higher, which significantly increase activations.
Large internal activations for backbone: our backbone uses a relatively large expansion ratio (6), causing the large expanded activations.
Deep BiFPN: our BiFPN has multiple top-down and bottom-up paths, which leads to a lot of intermediate memory usage during training.