Hello
I’m totally new to transfer learning.
I’m did transfer learning on a MobileNet-V1-SSD to detect strawberries in a picture.
I used plastic strawberries which are all identical.
My training data size is 424, test is around 50, validation around 15 (small dataset)
epochs 130
learning rate 0.01
t_max 100
basenet learning rate 0.001
scheduler cosine
The lowest loss was around 1.92
The strawberry pictures were taken of different sizes and angles and all on a white background.
The resulting model was very fast (on Jetson Nano) but it was taking pretty much everything for a strawberry (table, hand, head, cell phone… )
Since I’m completely new to this, I wonder if there is something obvious that I’m doing wrong?
Is there anything that looks wrong in what I’m doing? Too small dataset, white background… ? Poor parameters?
Thanks a lot!!
