no message

Author: Simon Niklaus

autozoom.py

@@ -47,18 +47,24 @@
 
 ##########################################################
 
-arguments_strIn = './images/doublestrike.jpg'
-arguments_strOut = './autozoom.mp4'
-
-for strOption, strArgument in getopt.getopt(sys.argv[1:], '', [ strParameter[2:] + '=' for strParameter in sys.argv[1::2] ])[0]:
-	if strOption == '--in' and strArgument != '': arguments_strIn = strArgument # path to the input image
-	if strOption == '--out' and strArgument != '': arguments_strOut = strArgument # path to where the output should be stored
+args_strIn = './images/doublestrike.jpg'
+args_strOut = './autozoom.mp4'
+args_strDepth = None
+
+for strOption, strArg in getopt.getopt(sys.argv[1:], '', [
+	'in=',
+	'out=',
+	'depth=',
+])[0]:
+	if strOption == '--in' and strArg != '': args_strIn = strArg # path to the input image
+	if strOption == '--out' and strArg != '': args_strOut = strArg # path to where the output should be stored
+	if strOption == '--depth' and strArg != '': args_strDepth = strArg # optional path to a depth map in numpy format
 # end
 
 ##########################################################
 
 if __name__ == '__main__':
-	npyImage = cv2.imread(filename=arguments_strIn, flags=cv2.IMREAD_COLOR)
+	npyImage = cv2.imread(filename=args_strIn, flags=cv2.IMREAD_COLOR)
 
 	intWidth = npyImage.shape[1]
 	intHeight = npyImage.shape[0]
@@ -70,7 +76,7 @@
 
 	npyImage = cv2.resize(src=npyImage, dsize=(intWidth, intHeight), fx=0.0, fy=0.0, interpolation=cv2.INTER_AREA)
 
-	process_load(npyImage, {})
+	process_load(npyImage, {} if args_strDepth is None else {'npyDepth': numpy.load(args_strDepth)})
 
 	objFrom = {
 		'fltCenterU': intWidth / 2.0,
@@ -92,5 +98,5 @@
 		'boolInpaint': True
 	})
 
-	moviepy.editor.ImageSequenceClip(sequence=[ npyFrame[:, :, ::-1] for npyFrame in npyResult + list(reversed(npyResult))[1:-1] ], fps=25).write_videofile(arguments_strOut)
+	moviepy.editor.ImageSequenceClip(sequence=[ npyFrame[:, :, ::-1] for npyFrame in npyResult + list(reversed(npyResult))[1:-1] ], fps=25).write_videofile(args_strOut)
 # end

common.py

@@ -5,11 +5,20 @@ def process_load(npyImage, objSettings):
 	objCommon['intHeight'] = npyImage.shape[0]
 
 	tenImage = torch.FloatTensor(numpy.ascontiguousarray(npyImage.transpose(2, 0, 1)[None, :, :, :].astype(numpy.float32) * (1.0 / 255.0))).cuda()
-	tenDisparity = disparity_estimation(tenImage)
-	tenDisparity = disparity_adjustment(tenImage, tenDisparity)
-	tenDisparity = disparity_refinement(tenImage, tenDisparity)
-	tenDisparity = tenDisparity / tenDisparity.max() * objCommon['fltBaseline']
-	tenDepth = (objCommon['fltFocal'] * objCommon['fltBaseline']) / (tenDisparity + 0.0000001)
+
+	if 'npyDepth' not in objSettings:
+		tenDisparity = disparity_estimation(tenImage)
+		tenDisparity = disparity_adjustment(tenImage, tenDisparity)
+		tenDisparity = disparity_refinement(tenImage, tenDisparity)
+		tenDisparity = tenDisparity / tenDisparity.max() * objCommon['fltBaseline']
+		tenDepth = (objCommon['fltFocal'] * objCommon['fltBaseline']) / (tenDisparity + 0.0000001)
+
+	elif 'npyDepth' in objSettings:
+		tenDepth = torch.FloatTensor(numpy.ascontiguousarray(numpy.atleast_3d(objSettings['npyDepth']).astype(numpy.float32).transpose(2, 0, 1)[None, :, :, :])).cuda()
+		tenDisparity = (objCommon['fltFocal'] * objCommon['fltBaseline']) / (tenDepth + 0.0000001)
+
+	# end
+
 	tenValid = (spatial_filter(tenDisparity / tenDisparity.max(), 'laplacian').abs() < 0.03).float()
 	tenPoints = depth_to_points(tenDepth * tenValid, objCommon['fltFocal'])
 	tenUnaltered = depth_to_points(tenDepth, objCommon['fltFocal'])
@@ -61,7 +70,7 @@ def process_shift(objSettings):
 	fltShiftY = fltClosestFromY - fltClosestToY
 	fltShiftZ = objSettings['fltDepthTo'] - objSettings['fltDepthFrom']
 
-	tenShift = torch.FloatTensor([ fltShiftX, fltShiftY, fltShiftZ ]).view(1, 3, 1).cuda()
+	tenShift = torch.tensor(data=[[[fltShiftX], [fltShiftY], [fltShiftZ]]], dtype=torch.float32, device=torch.device('cuda'))
 
 	tenPoints = objSettings['tenPoints'].clone()
 
@@ -140,7 +149,7 @@ def process_kenburns(objSettings):
 		objCommon['tenInpaDepth'] = objCommon['tenRawDepth'].view(1, 1, -1)
 		objCommon['tenInpaPoints'] = objCommon['tenRawPoints'].view(1, 3, -1)
 
-		for fltStep in [ 0.0, 1.0 ]:
+		for fltStep in [ 1.0 ]:
 			fltFrom = 1.0 - fltStep
 			fltTo = 1.0 - fltFrom
 

depthestim.py

@@ -47,18 +47,21 @@
 
 ##########################################################
 
-arguments_strIn = './images/doublestrike.jpg'
-arguments_strOut = './depthestim.npy'
-
-for strOption, strArgument in getopt.getopt(sys.argv[1:], '', [ strParameter[2:] + '=' for strParameter in sys.argv[1::2] ])[0]:
-	if strOption == '--in' and strArgument != '': arguments_strIn = strArgument # path to the input image
-	if strOption == '--out' and strArgument != '': arguments_strOut = strArgument # path to where the output should be stored
+args_strIn = './images/doublestrike.jpg'
+args_strOut = './depthestim.npy'
+
+for strOption, strArg in getopt.getopt(sys.argv[1:], '', [
+	'in=',
+	'out=',
+])[0]:
+	if strOption == '--in' and strArg != '': args_strIn = strArg # path to the input image
+	if strOption == '--out' and strArg != '': args_strOut = strArg # path to where the output should be stored
 # end
 
 ##########################################################
 
 if __name__ == '__main__':
-	npyImage = cv2.imread(filename=arguments_strIn, flags=cv2.IMREAD_COLOR)
+	npyImage = cv2.imread(filename=args_strIn, flags=cv2.IMREAD_COLOR)
 
 	fltFocal = max(npyImage.shape[1], npyImage.shape[0]) / 2.0
 	fltBaseline = 40.0
@@ -72,7 +75,7 @@
 	npyDisparity = tenDisparity[0, 0, :, :].cpu().numpy()
 	npyDepth = tenDepth[0, 0, :, :].cpu().numpy()
 
-	cv2.imwrite(filename=arguments_strOut.replace('.npy', '.png'), img=(npyDisparity / fltBaseline * 255.0).clip(0.0, 255.0).astype(numpy.uint8))
+	cv2.imwrite(filename=args_strOut.replace('.npy', '.png'), img=(npyDisparity / fltBaseline * 255.0).clip(0.0, 255.0).astype(numpy.uint8))
 
-	numpy.save(arguments_strOut, npyDepth)
+	numpy.save(args_strOut, npyDepth)
 # end

interface.py

@@ -84,7 +84,7 @@ def load_image():
 	process_load(objPlayback['npyImage'], {})
 
 	for fltX, fltY in [ (100.0, 0.0), (-100.0, 0.0), (0.0, 100.0), (0.0, -100.0) ]:
-		process_inpaint(torch.FloatTensor([ fltX, fltY, 0.0 ]).view(1, 3, 1).cuda())
+		process_inpaint(torch.tensor(data=[[[fltX], [fltY], [0.0]]], dtype=torch.float32, device=torch.device('cuda')))
 	# end
 
 	return ''