2
$\begingroup$

I am wondering why tf object detection api needs so few picture samples for training while regular cnns needs many more?

What I read in tutorials is that tf object detection api needs around 100-500 pictures per class for training (is it true?) while regular CNNs need many many more samples, like tens of thousands or more. Why is it so?

|improve this question
$\endgroup$
  • 1
    $\begingroup$ I haven't work with the tf object detection api so I might be totally wrong but I guess that they need so little data because their models are already trained on huge datasets, and they are just transferring the learning. $\endgroup$ – razvanc92 Aug 22 '19 at 14:09
  • $\begingroup$ You actually might be right :) haven't thought about it, thanks! $\endgroup$ – Makintosz Aug 22 '19 at 14:17
  • $\begingroup$ you are right. Actually from scratch OD needs more images than Classifiers because its a more difficult task $\endgroup$ – mshlis Aug 22 '19 at 15:26
  • 1
    $\begingroup$ @razvanc92 you can put your comment as an answer for this question :) $\endgroup$ – malioboro Aug 23 '19 at 8:54
2
$\begingroup$

I guess that they need so little data because their models are already trained on huge datasets, and they are just transferring the learning (using those pre-trained models as starting point).

|improve this answer
$\endgroup$

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.