Support for object detection using SqueezeDet
Hi,
Are there any future plans to extend the tf_importer so that it could also support SqueezeDet? At the moment, there are some layers missing (e.g. strided_slice) and I wouldn't really know how to implement these myself.
I find SqueezeDet one of the better architectures for object detection and it would be great if the graph models could be deployed through OpenCV.
Kind regards,
Milan
Hi, Milan! Please leave a reference to model to make sure that we'll test the right one.
Hi,
Thanks for the response!
I'm referring to a trained model (for the KITTI vision dataset) from this repository: https://github.com/BichenWuUCB/squeez... I've put aside the model of interest along with its frozen graph (which I'm trying to deploy in OpenCV) here: https://www.dropbox.com/sh/vxbf3g98cx...
It has the following layers: https://justpaste.it/1daoy (too much to paste here) The tensorflow importer parses everything up until "conv12/bias_add", but it has no functionality for some of the other layers/operations (which can't be stripped because they are needed to interpret the output).
Best regards,
Milan