def relative_marker_positions(im_pos, markers):
feature_map = np.zeros([im_pos.shape[0], im_pos.shape[1],
len(markers)],
dtype=np.float)
dgn.relative_distance_features(np.ascontiguousarray(im_pos),
np.ascontiguousarray(markers), feature_map)
return feature_map
def connect_extrema(im_pos, target, markers, visualize=False): ''' im_pos : XYZ positions of each point in image formation (n x m x 3) ''' height, width,_ = im_pos.shape centroid = np.array(target) im_pos = np.ascontiguousarray(im_pos.astype(np.int16)) cost_map = np.ascontiguousarray(np.zeros([height, width], dtype=np.uint16)) extrema = dgn.geodesic_map_MPI(cost_map, im_pos, np.array(centroid, dtype=np.int16), 1, 1) cost_map = extrema[-1] trails = [] for m in markers: trail = dgn.geodesic_trail(cost_map.copy()+(32000*(im_pos[:,:,2]==0)).astype(np.uint16), np.array(m, dtype=np.int16)) trails += [trail.copy()] if visualize: cost_map = deepcopy(cost_map) circ = circle(markers[0][0],markers[0][1], 5) circ = np.array([np.minimum(circ[0], height-1), np.minimum(circ[1], width-1)]) circ = np.array([np.maximum(circ[0], 0), np.maximum(circ[1], 0)]) cost_map[circ[0], circ[1]] = 0 for i,t in enumerate(trails[1:]): # embed() cost_map[t[:,0], t[:,1]] = 0 circ = circle(markers[i+1][0],markers[i+1][1], 5) circ = np.array([np.minimum(circ[0], height-1), np.minimum(circ[1], width-1)]) circ = np.array([np.maximum(circ[0], 0), np.maximum(circ[1], 0)]) cost_map[circ[0], circ[1]] = 0 return trails, cost_map else: return trails
def geodesic_extrema_MPI(im_pos,
centroid=None,
iterations=1,
visualize=False,
box=None):
'''
im : im_pos (NxMx3)
'''
if centroid == None:
try:
centroid = np.array(nd.center_of_mass(im_pos[:, :, 2]),
dtype=np.int16)
except:
return np.array([])
if box is not None:
im_pos = im_pos[box]
im_pos = np.ascontiguousarray(im_pos, dtype=np.int16)
if visualize:
cost_map = np.zeros([im_pos.shape[0], im_pos.shape[1]],
dtype=np.uint16)
extrema = dgn.geodesic_map_MPI(cost_map, im_pos,
np.array(centroid, dtype=np.int16),
iterations, 1)
cost_map = np.array(extrema[-1])
extrema = extrema[:-1]
extrema = np.array([x for x in extrema])
return extrema, cost_map
else:
extrema = np.array(
dgn.geodesic_extrema_MPI(im_pos, np.array(centroid,
dtype=np.int16),
iterations))
return extrema
def distance_map(im, centroid, scale=1):
'''
---Parameters---
im_depth :
centroid :
---Returns---
distance_map
'''
im_depth = np.ascontiguousarray(im.copy())
objSize = im_depth.shape
max_value = 32000
mask = im_depth > 0
# Get discrete form of position/depth matrix
# embed()
depth_min = im_depth[mask].min()
depth_max = im_depth[mask].max()
depth_diff = depth_max - depth_min
if depth_diff < 1:
depth_diff = 1
scale_to = scale / float(depth_diff)
# Ensure the centroid is within the boundaries
# Segfaults if on the very edge(!) so set border as 1 to resolution-2
centroid[0] = centroid[0] if centroid[0] > 0 else 1
centroid[
0] = centroid[0] if centroid[0] < im.shape[0] - 1 else im.shape[0] - 2
centroid[1] = centroid[1] if centroid[1] > 0 else 1
centroid[
1] = centroid[1] if centroid[1] < im.shape[1] - 1 else im.shape[1] - 2
# Scale depth image
im_depth_scaled = np.ascontiguousarray(
np.array((im_depth - depth_min) * scale_to, dtype=np.uint16))
# im_depth_scaled = np.ascontiguousarray(np.array( (im_depth-depth_min), dtype=np.uint16))
im_depth_scaled *= mask
# Initialize all but starting point as max
distance_map = np.zeros([objSize[0], objSize[1]],
dtype=np.uint16) + max_value
distance_map[centroid[0], centroid[1]] = 0
# Set which pixels are in/out of bounds
visited_map = np.zeros_like(distance_map, dtype=np.uint8)
visited_map[-mask] = 255
centroid = np.array(centroid, dtype=np.int16)
# embed()
dgn.distance_map(distance_map, visited_map,
im_depth_scaled.astype(np.uint16), centroid, int(scale))
return distance_map.copy()
def distance_map(im, centroid, scale=1): ''' ---Parameters--- im_depth : centroid : ---Returns--- distance_map ''' im_depth = np.ascontiguousarray(im.copy()) objSize = im_depth.shape max_value = 32000 mask = im_depth > 0 # Get discrete form of position/depth matrix # embed() depth_min = im_depth[mask].min() depth_max = im_depth[mask].max() depth_diff = depth_max - depth_min if depth_diff < 1: depth_diff = 1 scale_to = scale / float(depth_diff) # Ensure the centroid is within the boundaries # Segfaults if on the very edge(!) so set border as 1 to resolution-2 centroid[0] = centroid[0] if centroid[0] > 0 else 1 centroid[0] = centroid[0] if centroid[0] < im.shape[0]-1 else im.shape[0]-2 centroid[1] = centroid[1] if centroid[1] > 0 else 1 centroid[1] = centroid[1] if centroid[1] < im.shape[1]-1 else im.shape[1]-2 # Scale depth image im_depth_scaled = np.ascontiguousarray(np.array( (im_depth-depth_min)*scale_to, dtype=np.uint16)) # im_depth_scaled = np.ascontiguousarray(np.array( (im_depth-depth_min), dtype=np.uint16)) im_depth_scaled *= mask # Initialize all but starting point as max distance_map = np.zeros([objSize[0],objSize[1]], dtype=np.uint16)+max_value distance_map[centroid[0], centroid[1]] = 0 # Set which pixels are in/out of bounds visited_map = np.zeros_like(distance_map, dtype=np.uint8) visited_map[-mask] = 255 centroid = np.array(centroid, dtype=np.int16) # embed() dgn.distance_map(distance_map, visited_map, im_depth_scaled.astype(np.uint16), centroid, int(scale)) return distance_map.copy()
def connect_extrema(im_pos, target, markers, visualize=False):
'''
im_pos : XYZ positions of each point in image formation (n x m x 3)
'''
height, width, _ = im_pos.shape
centroid = np.array(target)
im_pos = np.ascontiguousarray(im_pos.astype(np.int16))
cost_map = np.ascontiguousarray(np.zeros([height, width], dtype=np.uint16))
extrema = dgn.geodesic_map_MPI(cost_map, im_pos,
np.array(centroid, dtype=np.int16), 1, 1)
cost_map = extrema[-1]
trails = []
for m in markers:
trail = dgn.geodesic_trail(
cost_map.copy() + (32000 *
(im_pos[:, :, 2] == 0)).astype(np.uint16),
np.array(m, dtype=np.int16))
trails += [trail.copy()]
if visualize:
cost_map = deepcopy(cost_map)
circ = circle(markers[0][0], markers[0][1], 5)
circ = np.array(
[np.minimum(circ[0], height - 1),
np.minimum(circ[1], width - 1)])
circ = np.array([np.maximum(circ[0], 0), np.maximum(circ[1], 0)])
cost_map[circ[0], circ[1]] = 0
for i, t in enumerate(trails[1:]):
# embed()
cost_map[t[:, 0], t[:, 1]] = 0
circ = circle(markers[i + 1][0], markers[i + 1][1], 5)
circ = np.array([
np.minimum(circ[0], height - 1),
np.minimum(circ[1], width - 1)
])
circ = np.array([np.maximum(circ[0], 0), np.maximum(circ[1], 0)])
cost_map[circ[0], circ[1]] = 0
return trails, cost_map
else:
return trails
def geodesic_extrema_MPI(im_pos, centroid=None, iterations=1, visualize=False, box=None): ''' im : im_pos (NxMx3) ''' if centroid==None: try: centroid = np.array(nd.center_of_mass(im_pos[:,:,2]), dtype=np.int16) except: return np.array([]) if box is not None: im_pos = im_pos[box] im_pos = np.ascontiguousarray(im_pos, dtype=np.int16) if visualize: cost_map = np.zeros([im_pos.shape[0], im_pos.shape[1]], dtype=np.uint16) extrema = dgn.geodesic_map_MPI(cost_map, im_pos, np.array(centroid, dtype=np.int16), iterations, 1) cost_map = np.array(extrema[-1]) extrema = extrema[:-1] extrema = np.array([x for x in extrema]) return extrema, cost_map else: extrema = np.array(dgn.geodesic_extrema_MPI(im_pos, np.array(centroid, dtype=np.int16), iterations)) return extrema
def local_histograms(im, n_bins=2, max_bound=255, patch_size=5):
output = dgn.local_histograms(np.ascontiguousarray(im), n_bins, patch_size,
max_bound)
return output
def local_histograms(im, n_bins=2, max_bound=255, patch_size=5): output = dgn.local_histograms(np.ascontiguousarray(im), n_bins, patch_size, max_bound) return output
def relative_marker_positions(im_pos, markers): feature_map = np.zeros([im_pos.shape[0], im_pos.shape[1], len(markers)], dtype=np.float) dgn.relative_distance_features(np.ascontiguousarray(im_pos), np.ascontiguousarray(markers), feature_map) return feature_map
def generateKeypoints(depthMat, iters=1, centroid=[], use_centroid=False): objSize = depthMat.shape # Get discrete form of position/depth matrix if len(objSize) == 3: depthMatMin = depthMat[depthMat[:,:,2]>0, 2].min() depthMatMax = depthMat[:,:,2].max() depthMatDiscrete = np.ascontiguousarray(np.array((1-(depthMat[:,:,2]-depthMatMin)/(depthMatMax-depthMatMin))*32000, dtype=np.uint16)) else: depthMatMin = depthMat[depthMat>0].min()# + 100 depthMatMax = depthMat[depthMat>0].max() # depthMatDiscrete = np.ascontiguousarray(np.array((1-(depthMat-depthMatMin)/(depthMatMax-depthMatMin))*32000, dtype=np.uint16)) depthMatDiscrete = np.ascontiguousarray(np.array(((depthMat.astype(np.float)-depthMatMin)/(depthMatMax-depthMatMin).astype(np.float))*32000, dtype=np.uint16)) # mask = np.all(depthMat != 0, 2) mask = depthMat > 0 depthMatDiscrete *= mask # Get starting position if centroid == []: if len(objSize) == 3: com = np.array(nd.center_of_mass(depthMat[:,:,2]), dtype=int) else: com = np.array(nd.center_of_mass(depthMat), dtype=int) startingPos = np.array([com[0], com[1]], dtype=np.int16) else: startingPos = np.array(centroid, dtype=np.int16) extrema = [] # extrema = [startingPos] trail = [] allTrails = [] combinedTrail = set() # Run dijkstras for i in xrange(iters): distsMat = np.zeros([objSize[0],objSize[1]], dtype=np.uint16)+32000 # Set starting point as zero weight distsMat[startingPos[0], startingPos[1]] = 0 # Set which pixels are in/out of bounds visitMat = np.zeros_like(distsMat, dtype=np.uint8) visitMat[-mask] = 255 # Set previous trails as zero distance for j in combinedTrail: distsMat[j[0], j[1]] = 0 # Dijkstras! trail = dgn.graph_dijkstras(distsMat, visitMat, depthMatDiscrete.astype(np.uint16), startingPos) distsMat *= mask * (distsMat<32000) # Add/merge trails allTrails.append(trail) for j in trail: if j[0] > 0 and j[1] > 0: combinedTrail.add((j[0], j[1])) extrema.append(np.array([trail[0][0], trail[0][1]], dtype=np.int16)) # Get new starting position if use_centroid: startingPos = extrema[0] else: startingPos = np.array([extrema[-1][0], extrema[-1][1]], dtype=np.int16) # print startingPos # Create a union of all trails if allTrails != []: for trails_i in allTrails: for i in trails_i: if i[0] > 0 and i[1] > 0: distsMat[i[0], i[1]] = distsMat.max() # Show trail on image for i in extrema: distsMat[i[0]-3:i[0]+4, i[1]-3:i[1]+4] = distsMat.max() return extrema, allTrails, distsMat
