if nh < newH:
im = cv2.resize(im_cpu, (newW, newH), interpolation=cv2.INTER_AREA)
else:
im = cv2.resize(im_cpu, (newW, newH), interpolation=cv2.INTER_LINEAR)
newEnvWidth = newW
newEnvHeight = newH
im = (np.transpose(im, [2, 0, 1]).astype(np.float32) /
255.0)[np.newaxis, :, :, :]
im = im / im.max()
imBatchSmall = Variable(torch.from_numpy(im**(2.2))).cuda()
renderLayer = models.renderingLayer(isCuda=opt.cuda,
imWidth=newEnvWidth,
imHeight=newEnvHeight,
fov=fov,
envWidth=opt.envWidth,
envHeight=opt.envHeight)
output2env = models.output2env(isCuda=opt.cuda,
envWidth=opt.envWidth,
envHeight=opt.envHeight,
SGNum=opt.SGNum)
########################################################
# Build the cascade network architecture #
albedoPreds, normalPreds, roughPreds, depthPreds = [], [], [], []
albedoBSPreds, roughBSPreds, depthBSPreds = [], [], []
envmapsPreds, envmapsPredImages, renderedPreds = [], [], []
cAlbedos = []
cLights = []
torch.load('{0}/normalRefs{1}_{2}.pth'.format(opt.modelRoot,
opt.cascadeLevel,
opt.epochId)))
roughRefs[0].load_state_dict(
torch.load('{0}/roughRefs{1}_{2}.pth'.format(opt.modelRoot,
opt.cascadeLevel,
opt.epochId)))
depthRefs[0].load_state_dict(
torch.load('{0}/depthRefs{1}_{2}.pth'.format(opt.modelRoot,
opt.cascadeLevel,
opt.epochId)))
envRefs[0].load_state_dict(
torch.load('{0}/envRefs{1}_{2}.pth'.format(opt.modelRoot, opt.cascadeLevel,
opt.epochId)))
renderLayer = models.renderingLayer(gpuId=opt.gpuId, isCuda=opt.cuda)
globIllu1to2.load_state_dict(
torch.load('{0}/globIllu1to2_{1}.pth'.format(opt.modelRootGlob,
opt.epochIdGlob)))
globIllu2to3.load_state_dict(
torch.load('{0}/globIllu2to3_{1}.pth'.format(opt.modelRootGlob,
opt.epochIdGlob)))
for param in encoderRefs[0].parameters():
param.requires_grad = False
for param in albedoRefs[0].parameters():
param.requires_grad = False
for param in normalRefs[0].parameters():
param.requires_grad = False
for param in roughRefs[0].parameters():
roughDecoder = models.decoder0(mode=2 )
depthDecoder = models.decoder0(mode=4 )
encoder1 = models.encoder0(cascadeLevel = 1 )
albedoDecoder1 = models.decoder0(mode=0 )
normalDecoder1 = models.decoder0(mode=1 )
roughDecoder1 = models.decoder0(mode=2 )
depthDecoder1 = models.decoder0(mode=4 )
lightEncoder = models.encoderLight(cascadeLevel = 0, SGNum = opt.SGNum )
axisDecoder = models.decoderLight(mode=0, SGNum = opt.SGNum )
lambDecoder = models.decoderLight(mode = 1, SGNum = opt.SGNum )
weightDecoder = models.decoderLight(mode = 2, SGNum = opt.SGNum )
renderLayer = models.renderingLayer(isCuda = opt.cuda,
imWidth=opt.envCol, imHeight=opt.envRow,
envWidth = opt.envWidth, envHeight = opt.envHeight)
output2env = models.output2env(isCuda = opt.cuda,
envWidth = opt.envWidth, envHeight = opt.envHeight, SGNum = opt.SGNum )
####################################################################
#########################################
encoder.load_state_dict(
torch.load('{0}/encoder{1}_{2}.pth'.format(opt.experimentBRDF0, 0,
opt.nepochBRDF0-1 ) ).state_dict() )
albedoDecoder.load_state_dict(
torch.load('{0}/albedo{1}_{2}.pth'.format(opt.experimentBRDF0, 0,
opt.nepochBRDF0-1 ) ).state_dict() )
normalDecoder.load_state_dict(
