Initial commit.

Initial commit.

Author: pableeto

BRDFNet/BRDFNetTraining.py

@@ -0,0 +1,69 @@
+[device]
+randomSeed = 23333
+
+[solver]
+SolverType = Adam
+
+;learning rate
+lr = 0.001	
+momentum = 0.9
+lrDecay = 0.0001
+batchSize = 64
+weightDecay = 0
+
+[stopping]
+;enter -1 to disable certain stopping critiria
+nMaxEpoch = -1
+nMaxIter = 250000
+
+[loop]
+;self-augment on/off
+renderLoop = 1
+
+;automatic compute ratio between labeled data/unlabeled data usage
+autoLoopRatio = 0
+
+;or manually set the ratio
+normalBatchLength = 1
+loopBatchLength = 3
+
+;how many iter/epoch before self-augment
+loopStartEpoch = -1
+loopStartIteration = 20000
+
+[network]
+NetworkType = Ratio
+Channal = Full
+BN = 1
+color = 0
+
+[dataset]
+NormalizeInput = 0
+dataset = /media/v-xil/New Volume/BRDF_envlight_CUDA/train_envlight/train_full.txt
+unlabelDataset = /media/v-xil/New Volume/BRDF_envlight_CUDA/train_envlight/train_full.txt
+testDataset = /media/v-xil/New Volume/BRDF_envlight_CUDA/test_envlight/test_full.txt
+
+;use of labeled data - defualt are corner data only
+;rest are left as unlabeled data
+albedoRange = 0,9
+specRange = 0,9
+roughnessRange = 0,14
+
+unlabellightfile = lvlist_1.txt
+unlabellightcondition = 0
+
+;test data, deafult is test on full test set
+testalbedoRange = 0,1,2,3,4,5,6,7,8
+testspecRange = 0,1,2,3,4,5,6,7,8
+testroughnessRange = 0,1,2,3,4,5,6,7,8,9,10,11,12,13
+
+;deprecated
+thetaRange = 0,1,2,3,4,5,6,7,8,9,10
+phiRange = 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14
+envLighting = 1
+
+[display]
+displayStep = 1000
+loopdisplayStep = 500
+checkpointStepIteration = 20000
+checkpointStepEpoch = -1

BRDFNet/BRDF_Net_Config.ini

@@ -0,0 +1,170 @@
+import sys, os
+working_path = os.path.dirname(os.path.realpath(__file__))
+root_path = os.path.dirname(working_path)
+sys.path.append(root_path + r'/Render')
+sys.path.append(root_path + r'/Utils')
+import numpy as np
+import math
+from utils import load_pfm, save_pfm, pfmFromBuffer, pfmToBuffer, make_dir, toLDR
+from operator import itemgetter
+import caffe
+
+from FastRendererCUDA import FastRenderEngine
+
+params = {}
+os.chdir(working_path)
+
+#dense sampled dataset, single channal
+albedoCnt = 10
+specCnt = 10
+roughnessCnt = 15
+thetaCnt = 15
+phiCnt = 15
+
+RenderOutput = r''
+
+with open('folderPath.txt', 'r') as f:
+    params['geometryPath'] = r'../Render/sphere.obj'
+    params['scriptRoot'] = r'../Utils'
+    params['outFolder'] = f.readline().strip()
+    params['envMapFolder'] = f.readline().strip()
+
+def idToBRDFid(id, aCnt, sCnt, rCnt):
+    aCnt = 10
+    sCnt = 10
+    rCnt = 15
+
+    aid = id / (sCnt * rCnt)
+    sid = (id % (sCnt * rCnt)) / rCnt
+    rid = id - aid * (sCnt * rCnt) - sid * rCnt
+
+    return aid, sid, rid
+
+def sampleCube(aid_list, sid_list, rid_list, full_list, dataset_name):
+    out_list = []
+    for a in aid_list:
+        for s in sid_list:
+            for r in rid_list:
+                for tid in range(0, thetaCnt):
+                    for pid in range(0, phiCnt):
+                        out_list.append('{}_{}_{}_{}_{}'.format(a, s, r, tid, pid))
+    unlabel_list = list(set(full_list) - set(out_list))
+    with open(RenderOutput + r'/train/train_{}.txt'.format(dataset_name), 'w') as f:
+        for x in out_list:
+            f.write(x)
+            f.write('\n')
+
+    with open(RenderOutput + r'/train/train_unlabel_{}.txt'.format(dataset_name), 'w') as f:
+        for x in unlabel_list:
+            f.write(x)
+            f.write('\n')
+
+def getLightTransList(numX, numY):
+    lightList = []
+    angleXList = []
+    angleYList = []
+#    np.random.seed(23333)
+    for i in range(0, numX):
+        for j in range(0, numY):
+            angleY = np.random.uniform(0.0, 360.0)
+            angleX = np.random.uniform(-30.0, 10.0)
+            lightList.append('r,0,1,0,{}/r,1,0,0,{}/end'.format(angleY, angleX))
+            angleXList.append(angleX)
+            angleYList.append(angleY)
+    return lightList, angleXList, angleYList
+
+
+if __name__ == '__main__':
+    gpuid = int(sys.argv[1])
+    out_root = sys.argv[2]
+
+    make_dir(out_root + r'/train_envlight')
+    make_dir(out_root + r'/test_envlight')
+
+    trainAlbedo = np.linspace(0.05, 1.0, albedoCnt)
+    trainSpec = np.linspace(0.05, 1.0, specCnt)
+    trainRoughness = np.exp(np.linspace(math.log(0.02), math.log(1.0), roughnessCnt))
+
+    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))
+
+    print('Rendering Training data...\n')
+    ftrain = open(out_root + r'/train_envlight/train_full.txt', 'w')
+    for aid, a in enumerate(trainAlbedo):
+        for sid, s in enumerate(trainSpec):
+            for rid, r in enumerate(trainRoughness):
+                lightMatrix = np.zeros((len(lightID), 9, 2))
+                print('...{}_{}_{}\n'.format(aid, sid, rid))
+            
+                trainCube[aid,sid,rid] = [a,s,r]
+                brdfFolder = out_root + r'/train_envlight/{}_{}_{}'.format(aid, sid, rid)
+                make_dir(brdfFolder)
+                                         
+                OnlineRender.SetAlbedoValue([a, a, a])
+                OnlineRender.SetSpecValue([s, s, s])
+                OnlineRender.SetRoughnessValue(r)
+
+                for lid, l in enumerate(lightID):
+                    OnlineRender.SetEnvLightByID(l+1)
+                    lightView, lightX, lightY = getLightTransList(3, 3)
+                    for vid, v in enumerate(lightView):                    
+                        OnlineRender.SetLightXform(lightX[vid], lightY[vid])
+                        img = OnlineRender.Render()
+                        save_pfm(brdfFolder + r'/{}_{}.pfm'.format(lid, vid), img)
+                        ftrain.write('{}_{}_{}_{}_{}\n'.format(aid, sid, rid, lid, vid))
+                        lightMatrix[lid, vid, 0] = lightX[vid]
+                        lightMatrix[lid, vid, 1] = lightY[vid]
+
+                np.savetxt(out_root + r'/train_envlight/lightMatrix_{}_{}_{}.txt'.format(aid, sid, rid), lightMatrix.flatten())
+
+    ftrain.close()
+    np.savetxt(out_root + r'/train_envlight/brdfcube.txt', trainCube.flatten())
+    print('Done.\n')
+
+    print('Rendering Test data...\n')
+    ftest = open(out_root + r'/test_envlight/test_full.txt', 'w')
+    for aid, a in enumerate(testAlbedo):
+        for sid, s in enumerate(testSpec):
+            for rid, r in enumerate(testRoughness):
+                print('...{}_{}_{}\n'.format(aid, sid, rid))
+                testCube[aid, sid, rid] = [a, s, r]
+                brdfFolder = out_root + r'/test_envlight/{}_{}_{}'.format(aid, sid, rid)#+ offSetAlbedo[gpuid], sid + offSetSpec[gpuid], rid)
+                make_dir(brdfFolder)
+            
+                OnlineRender.SetAlbedoValue([a, a, a])
+                OnlineRender.SetSpecValue([s, s, s])
+                OnlineRender.SetRoughnessValue(r)
+
+                for lid, l in enumerate(lightID):
+                    OnlineRender.SetEnvLightByID(l+1)#(envMapFolder + r'\{:04d}.pfm'.format(l+1), '')
+                    img = OnlineRender.Render()
+                    save_pfm(brdfFolder + r'/{}_{}.pfm'.format(lid, 0), img)
+                    ftest.write('{}_{}_{}_{}_{}\n'.format(aid, sid, rid, lid, 0))
+            
+    ftest.close()
+    np.savetxt(out_root + r'/test_envlight/brdfcube.txt', testCube.flatten())