0

I have used this link: https://github.com/albanie/wider2pascal to change Widerface dataset annotations to Pascalvoc format, because SSD network is originally written for PascalVoc dataset. Now we want to run SSD network. What should we do to change SSD to two class detection? The code is complicated and we are confused Which lines should be changed.

Please help us.

1 Answers1

0

This SSD implementation provides tools and a tutorial on how to use a trained SSD for transfer learning on your own dataset. The solution (or at least one possible solution, the tutorial explains multiple alternatives) is to sub-sample the weight tensors of the classifier layers of the trained SSD so that their shapes fits with the number of classes in your dataset.

Here is the tutorial.

Alex
  • 3,316
  • 4
  • 26
  • 52