why accuracy drops after converting with convert.py

[hasiffirdos]

python3 convert.py yolo_tiny.cfg final.weights output.h5 I run this command and weights converted successfully but when I call predict it gives the wrong object with 99% accuracy while it was giving true obj with darknet weights file. with darknet weights:

with conveted keras model:

thanks in advance.

[dreambit]

I have the same problem, i set same accuracy = 0.2 both on darknet and yolo.py but darknet detects more objects, much more.

[dreambit]

fixed by increasing max_boxes = 20 to 100

[kidboy-man]

Hi @dreambit I would like to know where's the part (line or file) that you changed and what's the effect on it? Thank you

[dreambit]

@kidboy-man hi, i do not remember, but i think u can search for files that match the string max_boxes = 20

[malik-anhar]

Hi @dreambit I'm facing the same issue too.. changing the max_boxes to 100 doesn't help in my case..