Open dzspb opened 7 years ago
SSD uses some specified new layers/function, e.g. normalization layer, permute layer, and detection output layer. Probably you need to re-implement some of those if you only care about inference. Training is more complicated as it involves many other stuff.
Hi! Thanks for sharing your work. Just wondering is it possible to convert your SSD model in caffe into tensorflow? Cause it's really convenient to build something in tensorflow sometimes. Thanks again!