Octavio Arriaga

Sorry Hugo I have been busy (thesis). I will try to review the PR as soon as possible since I have no previous experience with dlib. Thank you for your...

thank you @mirceaciu for your interest in the code, I didn't want to deal explicitly with the box offsets being bigger than the image; therefore, I just skip the frame...

@lantuzi @fferroni One problem with changing the image size is that you will not be able to load the pre-trained weights. Also you would need to change the **ground truth...

Hello @NegatioN, I am currently also working on implementing the "recent" version of SSD, as well as trying to incorporate smaller models. As far as I can tell from the...

@fferroni if you use the first layers of a pre-trained network and manage to connect it to the model here with PriorBoxes layers then yes, you shouldn't change the PriorBox...

Hello @sklin93, a possible problem that you could be having is that you are not pre-processing the box dimensions correctly. Have you looked at how is done in the `get_data_from_XML.py`?...

I also observe the same behavior. However, I was able to get a val loss of 1.4 after 20 epochs and afterwards the val loss started increasing.

I used the pre-trained weights provided in the README file, which I believe are the weights from an older original implementation in caffe.

@meetshah1995 I have trained SSD with only the VGG16 weights and it was overfiting after ~20 epochs my lowest validation loss was of 1.4. I believe that better results can...

Hello @KevinYuk, the file prior_boxes_ssd300.pkl contains 7308 bounding boxes. Every box contains x_min, y_min, x_max, y_max, variance_1, variance_2, variance_3, variance_4. This bounding boxes are referred in the paper as prior...