Andrey Rykov

@BenMacKenzie yep, it's for auto-configuring priors based on a layers' outputs.

@fferroni could you, please, provide more information about the error? E.g. what is your `model.summary()` and what do you get after `BBoxUtility.assign_boxes(...)`? I mean, what tensor has shape [16,7308,4] and...

@NegatioN I don't see any problem here. My code assumes, that there can be only one possible `min_size` for prior, that is true for author's architectures. So, in their `Caffe`...

@eric82714 probably, there is some problem in the path to the image? I have never met this before, sorry.

@nsknsl actually, the name of the layers was taken from the original implementation and differs from the names in the paper, as I can remember. Moreover, in this repo the...

@PKUWison they should be related to **s_min** and **s_max**, but in the original implementation they are just hard-coded. I think, that the authors experimented with these values, and in the...

@PKUWison you could use [loadmat fucntion from scipy](https://docs.scipy.org/doc/scipy/reference/generated/scipy.io.loadmat.html). Actually, you don't need to convert annotations to xml, I use a little bit more convenient (at least for me) format. It...

@KevinYuk yes, you're right. Forward pass should work with `Theano` backend also but I've never tested it. Training part is `Tensorflow` only because of the lack of some methods in...

@KevinYuk sorry, I forgot about `ssd_layers.py` and `ssd_utils.py`, I am afraid, you'll have to reimplement the necessary parts in `Theano`, in case of `ssd_layers.py` it should be simple. You just...

@ayushchopra96 hi! I believe, that in #15 quite a lot is explained. Try to read this issue and cite some parts, that are not clear, and I'll try to explain...