Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

No significant reduction in loss. #41

Open
piedcoderr opened this issue Jul 22, 2019 · 0 comments
Open

No significant reduction in loss. #41

piedcoderr opened this issue Jul 22, 2019 · 0 comments

Comments

@piedcoderr
Copy link

Last part of my tiny-yolo-voc-1c.cfg file:

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=125
activation=leaky

[convolutional]
size=1
stride=1
pad=1
filters=30
activation=linear

[region]
anchors = 1.08,1.19,  3.42,4.41,  6.63,11.38,  9.42,5.11,  16.62,10.52
bias_match=1
classes=1
coords=4
num=5
softmax=1
jitter=.2
rescore=1

object_scale=5
noobject_scale=1
class_scale=1
coord_scale=1

absolute=1
thresh = .5
random=1

Even after i trained for 6 hours on gpu the loss seems to reduce very slowly and it was above 90.
Here is a small sample after epoch 3 and epoch 16:

Right now the training goes like this:

Epoch 3:

step 100 - loss 105.87599182128906 - moving ave loss 105.92378855073041 step 101 - loss 105.90802001953125 - moving ave loss 105.92221169761049 step 102 - loss 105.96070861816406 - moving ave loss 105.92606138966585 step 103 - loss 105.91077423095703 - moving ave loss 105.92453267379497 step 104 - loss 105.89263916015625 - moving ave loss 105.9213433224311 step 105 - loss 105.79179382324219 - moving ave loss 105.9083883725122 step 106 - loss 105.97239685058594 - moving ave loss 105.91478922031959 step 107 - loss 105.93550109863281 - moving ave loss 105.91686040815092 step 108 - loss 105.82689666748047 - moving ave loss 105.90786403408389 step 109 - loss 105.89955139160156 - moving ave loss 105.90703276983565 step 110 - loss 105.97517395019531 - moving ave loss 105.91384688787163 step 111 - loss 105.94683074951172 - moving ave loss 105.91714527403565 Finish 3 epoch(es) step 112 - loss 105.88888549804688 - moving ave loss 105.91431929643677 step 113 - loss 105.77619934082031 - moving ave loss 105.90050730087513 step 114 - loss 106.05404663085938 - moving ave loss 105.91586123387356 step 115 - loss 105.96916198730469 - moving ave loss 105.92119130921667 step 116 - loss 105.88710021972656 - moving ave loss 105.91778220026767 step 117 - loss 105.81925201416016 - moving ave loss 105.90792918165691 step 118 - loss 105.8878173828125 - moving ave loss 105.90591800177248 step 119 - loss 105.89876556396484 - moving ave loss 105.90520275799172 step 120 - loss 105.86831665039062 - moving ave loss 105.90151414723161 step 121 - loss 105.82453918457031 - moving ave loss 105.89381665096549 step 122 - loss 105.72496795654297 - moving ave loss 105.87693178152324 step 123 - loss 105.88580322265625 - moving ave loss 105.87781892563655

Epoch 16:

step 592 - loss 104.63841247558594 - moving ave loss 104.63666866604511 Finish 16 epoch(es) step 593 - loss 104.696533203125 - moving ave loss 104.6426551197531 step 594 - loss 104.6519775390625 - moving ave loss 104.64358736168404 step 595 - loss 104.64488220214844 - moving ave loss 104.64371684573048 step 596 - loss 104.63302612304688 - moving ave loss 104.64264777346212 step 597 - loss 104.54460144042969 - moving ave loss 104.63284314015887 step 598 - loss 104.62289428710938 - moving ave loss 104.63184825485394 step 599 - loss 104.47380065917969 - moving ave loss 104.61604349528652 step 600 - loss 104.67867279052734 - moving ave loss 104.6223064248106 step 601 - loss 104.47022247314453 - moving ave loss 104.60709802964399 step 602 - loss 104.56214904785156 - moving ave loss 104.60260313146475 step 603 - loss 104.53447723388672 - moving ave loss 104.59579054170695 step 604 - loss 104.71623229980469 - moving ave loss 104.60783471751672 step 605 - loss 104.48993682861328 - moving ave loss 104.59604492862638 step 606 - loss 104.55252838134766 - moving ave loss 104.59169327389851 step 607 - loss 104.4090576171875 - moving ave loss 104.57342970822742 step 608 - loss 104.76617431640625 - moving ave loss 104.5927041690453 step 609 - loss 104.65669250488281 - moving ave loss 104.59910300262905 step 610 - loss 104.4439697265625 - moving ave loss 104.5835896750224 step 611 - loss 104.50250244140625 - moving ave loss 104.57548095166078 step 612 - loss 104.66641235351562 - moving ave loss 104.58457409184626 step 613 - loss 104.51042175292969 - moving ave loss 104.57715885795461 step 614 - loss 104.57950592041016 - moving ave loss 104.57739356420016 step 615 - loss 104.4698257446289 - moving ave loss 104.56663678224304 step 616 - loss 104.62855529785156 - moving ave loss 104.5728286338039 step 617 - loss 104.53482818603516 - moving ave loss 104.56902858902703 step 618 - loss 104.41508483886719 - moving ave loss 104.55363421401105 step 619 - loss 104.50767517089844 - moving ave loss 104.5490383096998

So As you can see its taking too much time and their is no significant progress.
I have also posted in stackoverflow #57143339

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant