3d from single images

1. factored3d

i'd never expected to see this come true

it is using edge boxes for the original segmentation proposals

(currently run from matlab, baaa) maybe we can use opencv's implementation, and make some online tool from it ;)

edge_detection = cv2.ximgproc.createStructuredEdgeDetection('model.yml.gz')
edges = edge_detection.detectEdges(np.float32(img_fine) / 255.0)

orimap = edge_detection.computeOrientation(edges)
edges = edge_detection.edgesNms(edges, orimap)

edge_boxes = cv2.ximgproc.createEdgeBoxes(0.65,0.75,minScore=0.03)
edge_boxes.setMaxBoxes(1000)
boxes = edge_boxes.getBoundingBoxes(edges, orimap)

2. Dense Depth

an encoder - decoder network, which can produce a depth map from a single 2d image.
(still struggling to get it into opencv from keras)