in eft/db_processing/AMT_sample_gen.py [0:0]
def RenderFitting_allSamples(inputDir_list, bVisSpinOriginal=False, bSampling= True, bRenderInit = False):
g_rootDirName = '/run/media/hjoo/disk/data/cocoPose3D_amt/0_SPIN/0_allSampleVis'
# metaDir_list ={}
# metaDir_list['coco'] ='/run/media/hjoo/disk/data/cocoPose3D_amt/0_SPIN/0_exemplarOutput/meta_coco_2dskeletons'
# metaDir_list['mpii'] = '/run/media/hjoo/disk/data/cocoPose3D_amt/0_SPIN/0_exemplarOutput/meta_mpii_2dskeletons'
# metaDir_list['lspet'] = '/run/media/hjoo/disk/data/cocoPose3D_amt/0_SPIN/0_exemplarOutput/meta_lspet_2dskeletons'
# # metaDir_list['pennaction'] = '/run/media/hjoo/disk/data/cocoPose3D_amt/0_SPIN/0_exemplarOutput/meta_lspet_2dskeletons'
imgDir ={}
imgDir['mpii'] = '/run/media/hjoo/disk/data/mpii_human_pose_v1/images'
imgDir['lsp'] = '/run/media/hjoo/disk/data/lsp_dataset_original/images/'
imgDir['lspet'] = '/run/media/hjoo/disk/data/lspet_dataset/images_highres'
imgDir['coco'] = '/run/media/hjoo/disk/data/coco/train2014'
imgDir['pennaction'] = '/run/media/hjoo/disk/data/Penn_Action/frames'
imgDir['panoptic'] = '/run/media/hjoo/disk/data/panoptic_mtc/a4_release/hdImgs'
totalFileList = []
for dbName in inputDir_list:
rootDirName_db = os.path.join(g_rootDirName, os.path.basename(inputDir_list[dbName]))
if bVisSpinOriginal:
glViewer.SetMeshColor('red')
else:
glViewer.SetMeshColor('blue')
if False== os.path.exists(rootDirName_db):
os.mkdir(rootDirName_db)
#Folder Name
imageFolder = os.path.join(rootDirName_db, 'images')
if not os.path.exists(imageFolder):
os.mkdir(imageFolder)
croppedimageFolder = os.path.join(rootDirName_db, 'croppedImage')
if not os.path.exists(croppedimageFolder):
os.mkdir(croppedimageFolder)
skelCroppedimageFolder_ = os.path.join(rootDirName_db, 'skelCroppedImage')
if not os.path.exists(skelCroppedimageFolder_):
os.mkdir(skelCroppedimageFolder_)
overlaidImageFolder = os.path.join(rootDirName_db, 'overlaid')
if not os.path.exists(overlaidImageFolder):
os.mkdir(overlaidImageFolder)
sideImageFolder = os.path.join(rootDirName_db, 'side')
if not os.path.exists(sideImageFolder):
os.mkdir(sideImageFolder)
mergedImageFolder = os.path.join(rootDirName_db, 'merged')
if not os.path.exists(mergedImageFolder):
os.mkdir(mergedImageFolder)
inputDir = inputDir_list[dbName]
fileList = listdir(inputDir)
for fileName in fileList:
if bRenderInit:
if ("_init" in fileName):
totalFileList.append( (dbName,fileName))
else:
if ("_init" in fileName) == False:
totalFileList.append( (dbName,fileName))
selectedSampleList = totalFileList
# #Generate Output
rejectedNum = 0
loss_keypoints_2d_list =[]
maxShape_list =[]
for idx, sample in enumerate(sorted(selectedSampleList)):
dbName = sample[0]
fileName = sample[1]
inputDir = inputDir_list[dbName]
fileFullPath = join(inputDir, fileName)
# if '098225941' in fileName:
# # print(data['loss_keypoints_2d'])
# # print(data['pred_shape'])
# here=0
# else:
# continue
print(fileFullPath)
with open(fileFullPath,'rb') as f:
data = pickle.load(f)
# loss_keypoints_2d_list.append(data['loss_keypoints_2d'])
# maxShape_list.append( np.max(abs(data['pred_shape'])))
# continue
if bRenderInit:
fileFullPath_newName = join(inputDir, fileName.replace("_init",""))
with open(fileFullPath_newName,'rb') as f:
datameta = pickle.load(f)
data['pred_camera'] = datameta['pred_camera']
if dbName!='panoptic':
if IsValid(data) == False:
rejectedNum+=1
print("Rejected: so far{}".format(rejectedNum))
continue
if IsValid_strict(data) == False:
rejectedNum+=1
print("Rejected: so far{}".format(rejectedNum))
continue
# #Load meta data
# metaDir = metaDir_list[dbName]
# meta_fileFullPath = join(metaDir, fileName)
# with open(meta_fileFullPath,'rb') as f:
# metadata = pickle.load(f)
# data['keypoint2d'] = metadata['keypoint2d']
imagePath = data['imageName'][0]
if dbName=='pennaction':
imgFullPath =os.path.join(imgDir[dbName], os.path.basename(os.path.dirname(imagePath)), os.path.basename(imagePath) )
elif dbName=='panoptic':
imgFullPath =os.path.join(imgDir[dbName], os.path.basename( os.path.dirname(os.path.dirname(imagePath))), os.path.basename(os.path.dirname(imagePath)), os.path.basename(imagePath) )
else:
# imgFullPath =os.path.join(imgDir, os.path.basename(imgFullPath) )
imgFullPath =os.path.join(imgDir[dbName], os.path.basename(imagePath) )
print(imgFullPath)
# outputFileName = '{0}-{1}-{2}.jpg'.format(dbName, os.path.basename(imgFullPath), idx )
outputFileName = imgFullPath.replace('/','-')
mergedImgFileName = '{0}/{1}'.format(mergedImageFolder,outputFileName )
if os.path.exists(mergedImgFileName):
continue
assert os.path.exists(imgFullPath)
rawImg = cv2.imread(imgFullPath)
scale = data['scale'][0]
center = data['center'][0]
croppedImg_highRes = crop(rawImg, center, scale,
[constants.IMG_RES*3, constants.IMG_RES*3])
#Crop image
croppedImg = crop(rawImg, center, scale,
[constants.IMG_RES, constants.IMG_RES])
validity = data['keypoint2d'][0,:,2] ==1
valid_keypoints = data['keypoint2d'][0,validity,:2]
min_pt = np.min(valid_keypoints, axis=0)
max_pt = np.max(valid_keypoints, axis=0)
bbox= [ min_pt[0], min_pt[1], max_pt[0] - min_pt[0], max_pt[1] - min_pt[1]]
bbox[0] = max(bbox[0]-20,0)
bbox[1] = max(bbox[1]-20,0)
bbox[2] = bbox[2]+20
bbox[3] = bbox[3]+20
rawImage_wSkel = viewer2D.Vis_Skeleton_2D_SPIN49(data['keypoint2d'][0,:,:2], pt2d_visibility=data['keypoint2d'][0,:,2], image= rawImg.copy())
rawImage_wSkel = viewer2D.Vis_Bbox(rawImage_wSkel,bbox)
croppedImg_highRes_withSkel = crop(rawImage_wSkel, center, scale,
[constants.IMG_RES*3, constants.IMG_RES*3])
if False:
viewer2D.ImShow(croppedImg_highRes,name='croppedImg_highRes', waitTime=1)
viewer2D.ImShow(croppedImg_highRes_withSkel,waitTime=1,name='croppedImg_highRes_withSkel')
viewer2D.ImShow(croppedImg,name='croppedImg', waitTime=1)
viewer2D.ImShow(rawImg,waitTime=0)
#Export Image
if True:
imgFilePath = os.path.join(imageFolder,outputFileName)
cv2.imwrite(imgFilePath,rawImg)
imgFilePath = os.path.join(croppedimageFolder,outputFileName)
cv2.imwrite(imgFilePath,croppedImg_highRes)
imgFilePath = os.path.join(skelCroppedimageFolder_,outputFileName)
cv2.imwrite(imgFilePath,croppedImg_highRes_withSkel)
#Vis overlaid Img
ours_betas = torch.from_numpy(data['pred_shape'])
ours_pose_rotmat = torch.from_numpy(data['pred_pose_rotmat'])
spin_betas = torch.from_numpy(data['opt_beta'])
spin_pose = torch.from_numpy(data['opt_pose'])
pred_camera_vis = data['pred_camera']
#Visualize SMPL output
# Note that gt_model_joints is different from gt_joints as it comes from SMPL
ours_output = smpl(betas=ours_betas, body_pose=ours_pose_rotmat[:,1:], global_orient=ours_pose_rotmat[:,0].unsqueeze(1), pose2rot=False )
ours_vertices = ours_output.vertices.detach().cpu().numpy()
ours_joints_3d = ours_output.joints.detach().cpu().numpy()
spin_output = smpl(betas=spin_betas, body_pose=spin_pose[:,3:], global_orient=spin_pose[:,:3])
spin_vertices = spin_output.vertices.detach().cpu().numpy()
spin_joints_3d = spin_output.joints.detach().cpu().numpy()
#Centering
if False:
spin_pelvis = (ours_joints_3d[:, 27:28,:3] + ours_joints_3d[:,28:29,:3]) / 2
ours_joints_3d[:,:,:3] = ours_joints_3d[:,:,:3] - spin_pelvis #centering
ours_vertices = ours_vertices - spin_pelvis
spin_model_pelvis = (spin_joints_3d[:, 27:28,:3] + spin_joints_3d[:,28:29,:3]) / 2
spin_joints_3d[:,:,:3] = spin_joints_3d[:,:,:3] - spin_model_pelvis #centering
spin_vertices = spin_vertices - spin_model_pelvis
b =0
############### Visualize Mesh ###############
pred_vert_vis = ours_vertices[b]
# meshVertVis = gt_vertices[b].detach().cpu().numpy()
# meshVertVis = meshVertVis-pelvis #centering
pred_vert_vis *=pred_camera_vis[b,0]
pred_vert_vis[:,0] += pred_camera_vis[b,1] #no need +1 (or 112). Rendernig has this offset already
pred_vert_vis[:,1] += pred_camera_vis[b,2] #no need +1 (or 112). Rendernig has this offset already
pred_vert_vis*=112
pred_meshes = {'ver': pred_vert_vis, 'f': smpl.faces}
opt_vert_vis = spin_vertices[b]
opt_vert_vis *=pred_camera_vis[b,0]
opt_vert_vis[:,0] += pred_camera_vis[b,1] #no need +1 (or 112). Rendernig has this offset already
opt_vert_vis[:,1] += pred_camera_vis[b,2] #no need +1 (or 112). Rendernig has this offset already
opt_vert_vis*=112
opt_meshes = {'ver': opt_vert_vis, 'f': smpl.faces}
# glViewer.setMeshData([pred_meshes, opt_meshes], bComputeNormal= True)
# glViewer.setMeshData([pred_meshes], bComputeNormal= True)
# glViewer.setMeshData([pred_meshes, opt_meshes], bComputeNormal= True)
if bVisSpinOriginal:
glViewer.setMeshData([opt_meshes], bComputeNormal= True)
else:
glViewer.setMeshData([pred_meshes], bComputeNormal= True)
glViewer.setBackgroundTexture(croppedImg)
glViewer.setWindowSize(croppedImg.shape[1]*3, croppedImg.shape[0]*3)
glViewer.SetOrthoCamera(True)
# glViewer.show(0)
# continue
if True:#g_bRenderFile: #Save to File
glViewer.SetNearPlane(500)
glViewer.setWindowSize(croppedImg.shape[1]*3, croppedImg.shape[0]*3)
glViewer.show_SMPL_cameraView(True)
overlaidImageFile = os.path.join(overlaidImageFolder,outputFileName)
rendered_img_source = g_renderDir + '/scene_00000000.jpg'
cmd = "cp -v {0} {1}".format(rendered_img_source, overlaidImageFile)
subprocess.call(cmd,shell=True)
glViewer.SetNearPlane(50)
glViewer.setWindowSize(croppedImg.shape[1]*3, croppedImg.shape[0]*3)
glViewer.show_SMPL_sideView(True)
sideImageFileName = os.path.join(sideImageFolder,outputFileName)
rendered_img_source = g_renderDir + '/scene_00000000.jpg'
cmd = "cp -v {0} {1}".format(rendered_img_source, sideImageFileName)
subprocess.call(cmd,shell=True)
else:
glViewer.SetNearPlane(50)
glViewer.setWindowSize(croppedImg.shape[1]*3, croppedImg.shape[0]*3)
glViewer.show_SMPL_sideView(True)
# glViewer.show(0)
#Merge images
if True:
# croppedImg_highRes + overlaidImageFile + sideImageFileName
overlaidImg = cv2.imread(overlaidImageFile)
sideImg = cv2.imread(sideImageFileName)
overlaidImg_vis = cv2.resize(overlaidImg, (croppedImg_highRes.shape[1], croppedImg_highRes.shape[0]))
sideImg_vis = cv2.resize(sideImg, (croppedImg_highRes.shape[1], croppedImg_highRes.shape[0]))
mergedImg = np.concatenate( (croppedImg_highRes, overlaidImg_vis, sideImg_vis), axis=1)
# viewer2D.ImShow(overlaidImg)
mergedImgFileName = '{0}/{1}'.format(mergedImageFolder,outputFileName )
#Write output here
maxShapeBeta = abs(torch.max( abs(ours_betas)).item())
finalKeyloss = data['loss_keypoints_2d']
reconErrorStr_raw = "m_beta {:.2f} | l_2d {:.8f}".format(maxShapeBeta, finalKeyloss)
mergedImg = cv2.putText(mergedImg,reconErrorStr_raw, (50,mergedImg.shape[0]-50), cv2.FONT_HERSHEY_PLAIN, 2, (255,255,0),2)
cv2.imwrite(mergedImgFileName, mergedImg)
viewer2D.Plot(loss_keypoints_2d_list)
viewer2D.Plot(maxShape_list)