PLEASE HELP ME ( High average loss)

pjreddie / darknet

Convolutional Neural Networks

http://pjreddie.com/darknet/

Other

25.82k stars 21.33k forks source link

PLEASE HELP ME ( High average loss) #702

Open veer345 opened 6 years ago

veer345 commented 6 years ago

yolo6 yolo4 yolo3 yolo2 yolo1 I have attached sccreenshots I have been trying from many days but same thing occurrs again and again please help me out average loss should be around0.06 but for my case it seems to be very high in rage of 200-300 please help me out i have to do college project if this doesn't run then i am busted

fvlntn commented 6 years ago

batch=2 subdivisions=2 ???

Put something like batch=64 subdivisions= 16 or 32 or 64 depending on your GPU.

After 700 iterations you only loaded 1400 images Which translates to 1400/64 = 22 real iterations with batch=64.

memon-aliraza commented 6 years ago

Hi, I have the same high avg loss issue but it is between ~3, ~4. By batch size is 64 and subdivisions=64. Here is the current situation:

4576: 3.999848, 3.889072 avg, 0.001000 rate, 52.965477 seconds, 292864 images Loaded: 0.000026 seconds

Can you let me know other possible causes!

IvanZheleznyakov commented 5 years ago

@veer345 hi, did u solve it? I have the same problem, my current avg loss is about 100 during a lot of iterations and it isnt decreasing ):