evaluate AP table VS train AP table is different 

Dear auth,
when i use same model, same dataset,
but my AP table before train (before train it eval first time) 
is not same with when i use evaluate.py ( use evaluate.py)
i think i will get same result
but i get different result.
