File size: 19,652 Bytes
04fbff5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 |
#!/usr/bin/env python
import getopt
import math
import numpy
import PIL
import PIL.Image
import sys
import torch
try:
from .correlation import correlation # the custom cost volume layer
except:
sys.path.insert(0, './correlation'); import correlation # you should consider upgrading python
# end
##########################################################
assert(int(str('').join(torch.__version__.split('.')[0:2])) >= 13) # requires at least pytorch version 1.3.0
torch.set_grad_enabled(False) # make sure to not compute gradients for computational performance
torch.backends.cudnn.enabled = True # make sure to use cudnn for computational performance
##########################################################
arguments_strModel = 'default' # 'default', or 'kitti', or 'sintel'
arguments_strOne = './images/one.png'
arguments_strTwo = './images/two.png'
arguments_strOut = './out.flo'
for strOption, strArgument in getopt.getopt(sys.argv[1:], '', [ strParameter[2:] + '=' for strParameter in sys.argv[1::2] ])[0]:
if strOption == '--model' and strArgument != '': arguments_strModel = strArgument # which model to use
if strOption == '--one' and strArgument != '': arguments_strOne = strArgument # path to the first frame
if strOption == '--two' and strArgument != '': arguments_strTwo = strArgument # path to the second frame
if strOption == '--out' and strArgument != '': arguments_strOut = strArgument # path to where the output should be stored
# end
##########################################################
backwarp_tenGrid = {}
def backwarp(tenInput, tenFlow):
if str(tenFlow.shape) not in backwarp_tenGrid:
tenHor = torch.linspace(-1.0 + (1.0 / tenFlow.shape[3]), 1.0 - (1.0 / tenFlow.shape[3]), tenFlow.shape[3]).view(1, 1, 1, -1).repeat(1, 1, tenFlow.shape[2], 1)
tenVer = torch.linspace(-1.0 + (1.0 / tenFlow.shape[2]), 1.0 - (1.0 / tenFlow.shape[2]), tenFlow.shape[2]).view(1, 1, -1, 1).repeat(1, 1, 1, tenFlow.shape[3])
backwarp_tenGrid[str(tenFlow.shape)] = torch.cat([ tenHor, tenVer ], 1).cuda()
# end
tenFlow = torch.cat([ tenFlow[:, 0:1, :, :] / ((tenInput.shape[3] - 1.0) / 2.0), tenFlow[:, 1:2, :, :] / ((tenInput.shape[2] - 1.0) / 2.0) ], 1)
return torch.nn.functional.grid_sample(input=tenInput, grid=(backwarp_tenGrid[str(tenFlow.shape)] + tenFlow).permute(0, 2, 3, 1), mode='bilinear', padding_mode='zeros', align_corners=False)
# end
##########################################################
class Network(torch.nn.Module):
def __init__(self):
super().__init__()
class Features(torch.nn.Module):
def __init__(self):
super().__init__()
self.netOne = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=3, out_channels=32, kernel_size=7, stride=1, padding=3),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1)
)
self.netTwo = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=32, out_channels=32, kernel_size=3, stride=2, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=32, out_channels=32, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=32, out_channels=32, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1)
)
self.netThr = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=32, out_channels=64, kernel_size=3, stride=2, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=64, out_channels=64, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1)
)
self.netFou = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=64, out_channels=96, kernel_size=3, stride=2, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=96, out_channels=96, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1)
)
self.netFiv = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=96, out_channels=128, kernel_size=3, stride=2, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1)
)
self.netSix = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=128, out_channels=192, kernel_size=3, stride=2, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1)
)
# end
def forward(self, tenInput):
tenOne = self.netOne(tenInput)
tenTwo = self.netTwo(tenOne)
tenThr = self.netThr(tenTwo)
tenFou = self.netFou(tenThr)
tenFiv = self.netFiv(tenFou)
tenSix = self.netSix(tenFiv)
return [ tenOne, tenTwo, tenThr, tenFou, tenFiv, tenSix ]
# end
# end
class Matching(torch.nn.Module):
def __init__(self, intLevel):
super().__init__()
self.fltBackwarp = [ 0.0, 0.0, 10.0, 5.0, 2.5, 1.25, 0.625 ][intLevel]
if intLevel != 2:
self.netFeat = torch.nn.Sequential()
elif intLevel == 2:
self.netFeat = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=32, out_channels=64, kernel_size=1, stride=1, padding=0),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1)
)
# end
if intLevel == 6:
self.netUpflow = None
elif intLevel != 6:
self.netUpflow = torch.nn.ConvTranspose2d(in_channels=2, out_channels=2, kernel_size=4, stride=2, padding=1, bias=False, groups=2)
# end
if intLevel >= 4:
self.netUpcorr = None
elif intLevel < 4:
self.netUpcorr = torch.nn.ConvTranspose2d(in_channels=49, out_channels=49, kernel_size=4, stride=2, padding=1, bias=False, groups=49)
# end
self.netMain = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=49, out_channels=128, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=128, out_channels=64, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=64, out_channels=32, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=32, out_channels=2, kernel_size=[ 0, 0, 7, 5, 5, 3, 3 ][intLevel], stride=1, padding=[ 0, 0, 3, 2, 2, 1, 1 ][intLevel])
)
# end
def forward(self, tenOne, tenTwo, tenFeaturesOne, tenFeaturesTwo, tenFlow):
tenFeaturesOne = self.netFeat(tenFeaturesOne)
tenFeaturesTwo = self.netFeat(tenFeaturesTwo)
if tenFlow is not None:
tenFlow = self.netUpflow(tenFlow)
# end
if tenFlow is not None:
tenFeaturesTwo = backwarp(tenInput=tenFeaturesTwo, tenFlow=tenFlow * self.fltBackwarp)
# end
if self.netUpcorr is None:
tenCorrelation = torch.nn.functional.leaky_relu(input=correlation.FunctionCorrelation(tenOne=tenFeaturesOne, tenTwo=tenFeaturesTwo, intStride=1), negative_slope=0.1, inplace=False)
elif self.netUpcorr is not None:
tenCorrelation = self.netUpcorr(torch.nn.functional.leaky_relu(input=correlation.FunctionCorrelation(tenOne=tenFeaturesOne, tenTwo=tenFeaturesTwo, intStride=2), negative_slope=0.1, inplace=False))
# end
return (tenFlow if tenFlow is not None else 0.0) + self.netMain(tenCorrelation)
# end
# end
class Subpixel(torch.nn.Module):
def __init__(self, intLevel):
super().__init__()
self.fltBackward = [ 0.0, 0.0, 10.0, 5.0, 2.5, 1.25, 0.625 ][intLevel]
if intLevel != 2:
self.netFeat = torch.nn.Sequential()
elif intLevel == 2:
self.netFeat = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=32, out_channels=64, kernel_size=1, stride=1, padding=0),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1)
)
# end
self.netMain = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=[ 0, 0, 130, 130, 194, 258, 386 ][intLevel], out_channels=128, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=128, out_channels=64, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=64, out_channels=32, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=32, out_channels=2, kernel_size=[ 0, 0, 7, 5, 5, 3, 3 ][intLevel], stride=1, padding=[ 0, 0, 3, 2, 2, 1, 1 ][intLevel])
)
# end
def forward(self, tenOne, tenTwo, tenFeaturesOne, tenFeaturesTwo, tenFlow):
tenFeaturesOne = self.netFeat(tenFeaturesOne)
tenFeaturesTwo = self.netFeat(tenFeaturesTwo)
if tenFlow is not None:
tenFeaturesTwo = backwarp(tenInput=tenFeaturesTwo, tenFlow=tenFlow * self.fltBackward)
# end
return (tenFlow if tenFlow is not None else 0.0) + self.netMain(torch.cat([ tenFeaturesOne, tenFeaturesTwo, tenFlow ], 1))
# end
# end
class Regularization(torch.nn.Module):
def __init__(self, intLevel):
super().__init__()
self.fltBackward = [ 0.0, 0.0, 10.0, 5.0, 2.5, 1.25, 0.625 ][intLevel]
self.intUnfold = [ 0, 0, 7, 5, 5, 3, 3 ][intLevel]
if intLevel >= 5:
self.netFeat = torch.nn.Sequential()
elif intLevel < 5:
self.netFeat = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=[ 0, 0, 32, 64, 96, 128, 192 ][intLevel], out_channels=128, kernel_size=1, stride=1, padding=0),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1)
)
# end
self.netMain = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=[ 0, 0, 131, 131, 131, 131, 195 ][intLevel], out_channels=128, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=128, out_channels=128, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=128, out_channels=64, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=64, out_channels=64, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=64, out_channels=32, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1),
torch.nn.Conv2d(in_channels=32, out_channels=32, kernel_size=3, stride=1, padding=1),
torch.nn.LeakyReLU(inplace=False, negative_slope=0.1)
)
if intLevel >= 5:
self.netDist = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=32, out_channels=[ 0, 0, 49, 25, 25, 9, 9 ][intLevel], kernel_size=[ 0, 0, 7, 5, 5, 3, 3 ][intLevel], stride=1, padding=[ 0, 0, 3, 2, 2, 1, 1 ][intLevel])
)
elif intLevel < 5:
self.netDist = torch.nn.Sequential(
torch.nn.Conv2d(in_channels=32, out_channels=[ 0, 0, 49, 25, 25, 9, 9 ][intLevel], kernel_size=([ 0, 0, 7, 5, 5, 3, 3 ][intLevel], 1), stride=1, padding=([ 0, 0, 3, 2, 2, 1, 1 ][intLevel], 0)),
torch.nn.Conv2d(in_channels=[ 0, 0, 49, 25, 25, 9, 9 ][intLevel], out_channels=[ 0, 0, 49, 25, 25, 9, 9 ][intLevel], kernel_size=(1, [ 0, 0, 7, 5, 5, 3, 3 ][intLevel]), stride=1, padding=(0, [ 0, 0, 3, 2, 2, 1, 1 ][intLevel]))
)
# end
self.netScaleX = torch.nn.Conv2d(in_channels=[ 0, 0, 49, 25, 25, 9, 9 ][intLevel], out_channels=1, kernel_size=1, stride=1, padding=0)
self.netScaleY = torch.nn.Conv2d(in_channels=[ 0, 0, 49, 25, 25, 9, 9 ][intLevel], out_channels=1, kernel_size=1, stride=1, padding=0)
# eny
def forward(self, tenOne, tenTwo, tenFeaturesOne, tenFeaturesTwo, tenFlow):
tenDifference = ((tenOne - backwarp(tenInput=tenTwo, tenFlow=tenFlow * self.fltBackward)) ** 2).sum(1, True).sqrt().detach()
tenDist = self.netDist(self.netMain(torch.cat([ tenDifference, tenFlow - tenFlow.view(tenFlow.shape[0], 2, -1).mean(2, True).view(tenFlow.shape[0], 2, 1, 1), self.netFeat(tenFeaturesOne) ], 1)))
tenDist = (tenDist ** 2).neg()
tenDist = (tenDist - tenDist.max(1, True)[0]).exp()
tenDivisor = tenDist.sum(1, True).reciprocal()
tenScaleX = self.netScaleX(tenDist * torch.nn.functional.unfold(input=tenFlow[:, 0:1, :, :], kernel_size=self.intUnfold, stride=1, padding=int((self.intUnfold - 1) / 2)).view_as(tenDist)) * tenDivisor
tenScaleY = self.netScaleY(tenDist * torch.nn.functional.unfold(input=tenFlow[:, 1:2, :, :], kernel_size=self.intUnfold, stride=1, padding=int((self.intUnfold - 1) / 2)).view_as(tenDist)) * tenDivisor
return torch.cat([ tenScaleX, tenScaleY ], 1)
# end
# end
self.netFeatures = Features()
self.netMatching = torch.nn.ModuleList([ Matching(intLevel) for intLevel in [ 2, 3, 4, 5, 6 ] ])
self.netSubpixel = torch.nn.ModuleList([ Subpixel(intLevel) for intLevel in [ 2, 3, 4, 5, 6 ] ])
self.netRegularization = torch.nn.ModuleList([ Regularization(intLevel) for intLevel in [ 2, 3, 4, 5, 6 ] ])
self.load_state_dict({ strKey.replace('module', 'net'): tenWeight for strKey, tenWeight in torch.hub.load_state_dict_from_url(url='http://content.sniklaus.com/github/pytorch-liteflownet/network-' + arguments_strModel + '.pytorch').items() })
# self.load_state_dict(torch.load('./liteflownet/network-default.pth'))
# end
def forward(self, tenOne, tenTwo):
tenOne[:, 0, :, :] = tenOne[:, 0, :, :] - 0.411618
tenOne[:, 1, :, :] = tenOne[:, 1, :, :] - 0.434631
tenOne[:, 2, :, :] = tenOne[:, 2, :, :] - 0.454253
tenTwo[:, 0, :, :] = tenTwo[:, 0, :, :] - 0.410782
tenTwo[:, 1, :, :] = tenTwo[:, 1, :, :] - 0.433645
tenTwo[:, 2, :, :] = tenTwo[:, 2, :, :] - 0.452793
tenFeaturesOne = self.netFeatures(tenOne)
tenFeaturesTwo = self.netFeatures(tenTwo)
tenOne = [ tenOne ]
tenTwo = [ tenTwo ]
for intLevel in [ 1, 2, 3, 4, 5 ]:
tenOne.append(torch.nn.functional.interpolate(input=tenOne[-1], size=(tenFeaturesOne[intLevel].shape[2], tenFeaturesOne[intLevel].shape[3]), mode='bilinear', align_corners=False))
tenTwo.append(torch.nn.functional.interpolate(input=tenTwo[-1], size=(tenFeaturesTwo[intLevel].shape[2], tenFeaturesTwo[intLevel].shape[3]), mode='bilinear', align_corners=False))
# end
tenFlow = None
for intLevel in [ -1, -2, -3, -4, -5 ]:
tenFlow = self.netMatching[intLevel](tenOne[intLevel], tenTwo[intLevel], tenFeaturesOne[intLevel], tenFeaturesTwo[intLevel], tenFlow)
tenFlow = self.netSubpixel[intLevel](tenOne[intLevel], tenTwo[intLevel], tenFeaturesOne[intLevel], tenFeaturesTwo[intLevel], tenFlow)
tenFlow = self.netRegularization[intLevel](tenOne[intLevel], tenTwo[intLevel], tenFeaturesOne[intLevel], tenFeaturesTwo[intLevel], tenFlow)
# end
return tenFlow * 20.0
# end
# end
netNetwork = None
##########################################################
def estimate(tenOne, tenTwo):
global netNetwork
if netNetwork is None:
netNetwork = Network().cuda().eval()
# end
assert(tenOne.shape[1] == tenTwo.shape[1])
assert(tenOne.shape[2] == tenTwo.shape[2])
intWidth = tenOne.shape[2]
intHeight = tenOne.shape[1]
# assert(intWidth == 1024) # remember that there is no guarantee for correctness, comment this line out if you acknowledge this and want to continue
# assert(intHeight == 436) # remember that there is no guarantee for correctness, comment this line out if you acknowledge this and want to continue
tenPreprocessedOne = tenOne.cuda().view(1, 3, intHeight, intWidth)
tenPreprocessedTwo = tenTwo.cuda().view(1, 3, intHeight, intWidth)
intPreprocessedWidth = int(math.floor(math.ceil(intWidth / 32.0) * 32.0))
intPreprocessedHeight = int(math.floor(math.ceil(intHeight / 32.0) * 32.0))
tenPreprocessedOne = torch.nn.functional.interpolate(input=tenPreprocessedOne, size=(intPreprocessedHeight, intPreprocessedWidth), mode='bilinear', align_corners=False)
tenPreprocessedTwo = torch.nn.functional.interpolate(input=tenPreprocessedTwo, size=(intPreprocessedHeight, intPreprocessedWidth), mode='bilinear', align_corners=False)
tenFlow = torch.nn.functional.interpolate(input=netNetwork(tenPreprocessedOne, tenPreprocessedTwo), size=(intHeight, intWidth), mode='bilinear', align_corners=False)
tenFlow[:, 0, :, :] *= float(intWidth) / float(intPreprocessedWidth)
tenFlow[:, 1, :, :] *= float(intHeight) / float(intPreprocessedHeight)
return tenFlow[0, :, :, :].cpu()
# end
##########################################################
if __name__ == '__main__':
tenOne = torch.FloatTensor(numpy.ascontiguousarray(numpy.array(PIL.Image.open(arguments_strOne))[:, :, ::-1].transpose(2, 0, 1).astype(numpy.float32) * (1.0 / 255.0)))
tenTwo = torch.FloatTensor(numpy.ascontiguousarray(numpy.array(PIL.Image.open(arguments_strTwo))[:, :, ::-1].transpose(2, 0, 1).astype(numpy.float32) * (1.0 / 255.0)))
tenOutput = estimate(tenOne, tenTwo)
objOutput = open(arguments_strOut, 'wb')
numpy.array([ 80, 73, 69, 72 ], numpy.uint8).tofile(objOutput)
numpy.array([ tenOutput.shape[2], tenOutput.shape[1] ], numpy.int32).tofile(objOutput)
numpy.array(tenOutput.numpy().transpose(1, 2, 0), numpy.float32).tofile(objOutput)
objOutput.close()
# end |