i am training yolov7 tiny with learning rate lr 0.01 and adam optimizer the data has class imbalance in two of the classes with total 3 classes i also applied cbam module in backbone but model is not improving accuracy and stuck at 20% or 10% accuracy map0.95.

yolov7-tiny on hpc via ssh

i was training image dataset of underwater trash with size of 9000+ with yolov7-tiny
hyper parameters were set to 0.01 lr initial learning rate and Adam as optimizer
cbam module is also applied in backbone
i was expecting that accuracy will get increase but no it remains stuck at 10 or 20 %