Pedestrian Detection with Caffe

[Miguel Pereira]

Hi Everyone,

I'm trying to gauge how feasible it would be to do some Pedestrian Detection experiments with caffe on Windows. I'm just starting out and I'm new to Caffe and machine learning in general. Could anyone help answer any of the following questions: - any help, hints or pointers would be greatly appreciated.  

1) I see RCNN can only use Caffe v0.999. Is there an easy way to install this version of caffe on a windows machine? 

2) Would I be better off starting with Caffe2 and Detectron? How accessible is Caffe2 anyway?  

3) The detect.py example on the latest Caffe version (i.e. clone the repo, checkout windows) uses sergeyk's selective_search. Doesn't this only run on python 2.7? 

4) Is it still feasible to use the pure Caffe implementation of RCNN with the latest caffe version? 

5) This is a long shot: Anyone have any information on the best way to use a Caffe CNN for classifying object proposals (specified using bounding boxes) for pedestrian detection? 

Many of the papers on pedestrian detection with CNNs use Caffe and are from 2014-2015, hence all of my questions. I'd like to avoid running into compatibility issues wherever possible. Any help would be greatly appreciated!