loader=jinja2.FileSystemLoader('./')).get_template('template.html')
savedNet = sys.argv[1]
testParamPath = sys.argv[2]
gpuid = int(sys.argv[3])
resultfolder, modelfile = os.path.split(savedNet)
#Load test params
test_params = {}
test_params = loadParams(testParamPath)
outputFolder = resultfolder + r'/test_{}'.format(test_params['outtag'])
if (os.path.exists(outputFolder) == False):
os.makedirs(outputFolder)
OnlineRender = FastRenderEngine(0)
OnlineRender.SetGeometry('Sphere') #params['geometryPath'], True, '')
OnlineRender.PreLoadAllLight(r'{}/light.txt'.format(
params['envMapFolder']))
fovRadian = 60.0 / 180.0 * math.pi
cameraDist = 1.5 / (math.tan(fovRadian / 2.0))
OnlineRender.SetCamera(0, 0, cameraDist, 0, 0, 0, 0, 1, 0, fovRadian, 0.01,
100, 128, 128)
OnlineRender.SetSampleCount(128, 512)
renderContext['experimentTag'] = savedNet
random.seed(23333)
np.random.seed(23333)
caffe.set_random_seed(23333)
if(len(sys.argv) == 9):
out_root = sys.argv[8]
else:
out_root = data_root
labeled_file_in = data_root + r'/{}/Labeled/trainingdata.txt'.format(data_tag)
test_file_in = data_root + r'/{}/Test/test.txt'.format(data_tag)
rendered_labeled_out = out_root + r'/{}/Labeled'.format(data_tag)
rendered_test_out = out_root + r'/{}/Test'.format(data_tag)
specular_file = data_root + r'/{}/Labeled/specroughness.txt'.format(data_tag)
lightpool_file = params_global['envMapFolder'] + r'/lightPool_{}.dat'.format(data_tag)
AugmentRender = FastRenderEngine(gpuid)
AugmentRender.SetGeometry('Plane')
AugmentRender.PreLoadAllLight(r'{}/light.txt'.format(params_global['envMapFolder']))
AugmentRender.SetSampleCount(128, 1024)
fovRadian = 60.0 / 180.0 * math.pi
cameraDist = 1.0 / (math.tan(fovRadian / 2.0))
AugmentRender.SetCamera(0, 0, cameraDist, 0, 0, 0, 0, 1, 0, fovRadian, 0.01, 100, 384, 384)
specList_final = {}
roughnessList_final = {}
with open(specular_file, 'r') as f:
rawList = f.read().strip().split('\n')
for t in rawList:
mid = int(t.split(',')[0])
spec = float(t.split(',')[1])
roughness = float(t.split(',')[2])
testAlbedo = np.linspace(0.1, 0.95, albedoCnt - 1)
testSpec = np.linspace(0.1, 0.95, specCnt - 1)
testRoughness = np.exp(
np.linspace(math.log(0.03), math.log(0.87), roughnessCnt - 1))
imageCnt = albedoCnt * specCnt * roughnessCnt * thetaCnt * phiCnt
envMapFolder = params['envMapFolder']
with open(envMapFolder + '/light.txt', 'r') as f:
lightID = map(int, f.read().strip().split('\n'))
lightID = list(np.array(lightID) - 1)
np.random.seed(gpuid)
OnlineRender = FastRenderEngine(gpuid)
OnlineRender.SetGeometry('Sphere')
OnlineRender.PreLoadAllLight(r'{}/light.txt'.format(envMapFolder))
fovRadian = 60.0 / 180.0 * math.pi
cameraDist = 1.5 / (math.tan(fovRadian / 2.0))
OnlineRender.SetCamera(0, 0, cameraDist, 0, 0, 0, 0, 1, 0, fovRadian, 0.01,
100, 128, 128)
OnlineRender.SetSampleCount(128, 1024)
OnlineRender.SetRenderMode(0)
albedoCnt = len(trainAlbedo)
specCnt = len(trainSpec)
roughnessCnt = len(trainRoughness)
trainCube = np.zeros((albedoCnt, specCnt, roughnessCnt, 3))
testCube = np.zeros((albedoCnt - 1, specCnt - 1, roughnessCnt - 1, 3))
imgbatch[i, :, :, :] = autoExposure(imgbatch[i, :, :, :])
imgbatch = imgbatch.transpose([0,2,3,1])
pipe.send(imgbatch)
counter = counter + batchSize
posInDataSet = (posInDataSet + batchSize) % dataset.dataSize
newepoch = counter / dataset.dataSize
if(newepoch != epoch):
dataset.shuffle()
epoch = newepoch
# init fast online renderer
onlineRender = FastRenderEngine(params['rendergpuid'])
onlineRender.SetGeometry('Plane')
onlineRender.SetSampleCount(128, 512)
onlineRender.PreLoadAllLight(r'{}/light.txt'.format(params['envMapFolder']))
fovRadian = 60.0 / 180.0 * np.pi
cameraDist = 1.0 / (math.tan(fovRadian / 2.0))
onlineRender.SetCamera(0, 0, cameraDist, 0, 0, 0, 0, 1, 0, fovRadian, 0.01, 100, 256, 256)
# set random seeds
random.seed(params['randomSeed'])
tf.set_random_seed(params['randomSeed'])
np.random.seed(params['randomSeed'])
# init tensorflow network
for id, lid in enumerate(lightID):
lightIDToEnumerateID[lid] = id
if __name__ == '__main__':
modelFile = sys.argv[1]
testSetPath = sys.argv[2]
gpuid = int(sys.argv[3])
imgw = 256
fovRadian = 60.0 / 180.0 * math.pi
cameraDist = 1.0 / (math.tan(fovRadian / 2.0))
cameraDist_1 = 1.5 / (math.tan(fovRadian / 2.0))
RelightingRender = FastRenderEngine(gpuid)
RelightingRender.SetGeometry('Plane')
RelightingRender.SetCamera(0, 0, cameraDist, 0, 0, 0, 0, 1, 0, fovRadian,
0.01, 100, 256, 256)
RelightingRender.SetSampleCount(128, 512)
RelightingRender.PreLoadAllLight(r'{}/light.txt'.format(
params_global['envMapFolder']))
caffe.set_mode_gpu()
caffe.set_device(gpuid)
path, file = os.path.split(modelFile)
modelFolder = path
testnet = caffe.Net(path + r'/net_test.prototxt', caffe.TEST)
testnet.copy_from(modelFile)
path, file = os.path.split(testSetPath)
with open(testSetPath, 'r') as f: