bad precision F1 and Recall accuracy

[copper carbon]

hello currently training 2 models (res50 and densenet121) the issue is i have good training and testing accuracy (94 for res and den 90) but my f1 score is so bad like we talking 1-4% during training duration, i'm doing image classifcation, on the amazon berkley object dataset, if u have any idea what to do or what can be the leading cause for this, comment below, thank you!,

[noble galleon]

Is your dataset imbalanced?

F1 score and accuracy measure entirely different things and each metric comes with their own inductive biases.

That being said, whatever you do choose.. Try to justify it.