1
$\begingroup$

I am using total of 3000 images for training an ssd_inception_v2_coco as the object detection model. I have set batch size as 4 because I don't have a high end GPU hence I am renting it for few hours and wanted to run the network for 40000 steps.
Till now the network has reached 15000 steps but the loss is very huge (4.12).

My questions are:

  1. Is the high loss because of the low batch size?
  2. If I am using a batch size very low then will it take like 100k steps to converge?
$\endgroup$
1
$\begingroup$

Irrespectively of training the model in a more general sense, you are likely checking on your model's far too often to spot any actual performance increase.

More specifically, if performance metrics are being printed out every 5 batches for example, you are seeing a performance increase based on just 5 x 4 = 20 data samples which is way too low for a DL model to have learnt anything meaningful. In the case of a batch_size = 256 the model would have seen $1260$ more samples than in your case.

I hope this helps.

$\endgroup$

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.