0
$\begingroup$

I am training YOLO network consisting of resnet50 architecture.This problem is to find different text labels on the image and predict bounding boxes

During training, I am seeing very less change in both training and validation loss. What are different method / debugging techniques to know where exacly is the problem. I am training on 5000 images and have around 23M parameters to train. I am using batch size as 8 and training times as 5 and number of epochs as 50.

enter image description here

$\endgroup$
0
$\begingroup$

Welcome to the site! My first thought is that 5k images is definitely not enough to hope to optimize 23m parameters, which is why your loss does not decrease.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ Ok, data is not a problem for me as I am generating it myself. could you give sense of how much images i should go for to train 23M parameters.To give more information , I have 9 classes to predict on each image. $\endgroup$ – vaibhav bansal Sep 21 '18 at 10:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.