loss_train = torch.nn.MSELoss()
reduction = 'none'
loss_test = losses_poses.MPJPECriterion(weight=1, reduction=reduction)
loss_train = losses_poses.StaticDenormalizedLoss(pose_key, loss_train)
loss_test = losses_generic.LossLabelMeanStdUnNormalized(
pose_key, loss_test, scale_normalized=False)
#if config_dict['spatial_transformer_num'] >= 2:
# loss_train = losses_poses.MultiPersonSimpleLoss(loss_train) # minimum of two
# loss_test = losses_poses.MultiPersonSimpleLoss(loss_test) # minimum of two
# annotation and pred is organized as a list, to facilitate multiple output types (e.g. heatmap and 3d loss)
return loss_train, loss_test
def get_parameter_description(self, config_dict): #, config_dict):
folder = "../output/trainPose_{note}_layers{num_encoding_layers}_implR{implicit_rotation}_fg{latent_fg}_3d{skip_background}_ldrop{latent_dropout}_billin{upsampling_bilinear}_fscale{feature_scale}_shuffleFG{shuffle_fg}_shuffle3d{shuffle_3d}_nth{every_nth_frame}_c{active_cameras}_sub{actor_subset}_bs{useCamBatches}_lr{learning_rate}_".format(
**config_dict)
folder = folder.replace(' ',
'').replace('../', '[DOT_SHLASH]').replace(
'.', 'o').replace('[DOT_SHLASH]',
'../').replace(',', '_')
return folder
if __name__ == "__main__":
config_dict_module = utils_io.loadModule(
"configs/config_train_detect_encode_decode_pose.py")
config_dict = config_dict_module.config_dict
ignite = IgniteTrainPose()
ignite.run(config_dict_module.__file__, config_dict)
utils_train.transfer_partial_weights(pretrained_states, network_single.to_pose, submodule=0) # last argument is to remove "network.single" prefix in saved network
print("Done loading weights from config_dict['pretrained_posenet_network_path']")
return network_single
def predict():
#nonlocal output_dict
#model.eval()
with torch.no_grad():
input_dict_cuda, label_dict_cuda = utils_data.nestedDictToDevice((input_dict, label_dict), device=device)
output_dict_cuda = model(input_dict_cuda)
output_dict = utils_data.nestedDictToDevice(output_dict_cuda, device='cpu')
if __name__ == "__main__":
params = param.get_general_params()
config_dict_module = utils_io.loadModule("configs/config_test_encodeDecode.py")
config_dict = config_dict_module.config_dict
config_dict['batch_size_test'] = 2
config_dict['n_hidden_to3Dpose'] = config_dict.get('n_hidden_to3Dpose', 2)
# load data
device='cuda'
data_loader = pickle.load(open('examples/test_set.pickl',"rb"))
ex = data_loader[0][0]
# load model
model = load_network(config_dict)
model = model.to(device)
from utils import io as utils_io
import os
config_dict = utils_io.loadModule(
"./configs/config_train_encodeDecode_pose.py").config_dict
config_dict['num_workers'] = 0
config_dict['label_types_test'].remove('img_crop')
config_dict['label_types_train'].remove('img_crop')
config_dict['batch_size_train'] = 1
config_dict['batch_size_test'] = 1
if 0:
network_path = '../output/trainNVS_resL3_ResNet_layers4_implRFalse_s3Dp[1_9_11]_w3Dp0_w3D0_wRGB1_wGrad0_wImgNet2_skipBG0_fg128_3dTrue_lh3Dp2_ldrop0o3_billinupper_fscale2_shuffleFGTrue_shuffle3dTrue_h36m_cross_nth1_cFalse_sub[1_5_6_7_8]_bs2_lr0o001_'
config_dict[
'pretrained_network_path'] = network_path + '/models/network_best_val_t1.pth'
else:
network_path = '../examples'
config_dict[
'pretrained_network_path'] = network_path + '/network_best_val_t1.pth'
if not os.path.exists(config_dict['pretrained_network_path']):
import urllib.request
print("Downloading pre-trained weights, can take a while...")
urllib.request.urlretrieve(
"http://documents.epfl.ch/groups/c/cv/cvlab-unit/www/data/ECCV2018Rhodin/network_best_val_t1.pth",
config_dict['pretrained_network_path'])
print("Downloading done.")
from utils import io as utils_io
#network_path = '/cvlabdata1/home/rhodin/code/humanposeannotation/python/pytorch_selfsupervised_multiview/output/encode_resL3_ResNet_layers4_implRFalse_s3Dp[1_9_11]_w3Dp0_w3D0_wRGB1_wGrad0_wImgNet2_skipBG0_fg128_3dTrue_lh3Dp2_ldrop0o3_billinupper_fscale2_shuffleFGTrue_shuffle3dTrue_h36m_cross_nth1_cFalse_sub[1_5_6_7_8]_bs2_lr0o001_'
if 0:
network_path = './output/trainNVS_resL3_ResNet_layers4_implRFalse_s3Dp[1_9_11]_w3Dp0_w3D0_wRGB1_wGrad0_wImgNet2_skipBG0_fg128_3dTrue_lh3Dp2_ldrop0o3_billinupper_fscale2_shuffleFGTrue_shuffle3dTrue_h36m_cross_nth1_cFalse_sub[1_5_6_7_8]_bs2_lr0o001_'
config_dict = utils_io.loadModule(
network_path + "/config_train_encodeDecode.py").config_dict
config_dict[
'pretrained_network_path'] = network_path + '/models/network_last_val.pth'
else:
import os
config_dict = utils_io.loadModule(
"configs/config_train_encodeDecode.py").config_dict
config_dict[
'pretrained_network_path'] = './examples/network_best_val_t1.pth'
if not os.path.exists(config_dict['pretrained_network_path']):
import urllib.request
print("Downloading pre-trained weights, can take a while...")
urllib.request.urlretrieve(
"http://documents.epfl.ch/groups/c/cv/cvlab-unit/www/data/ECCV2018Rhodin/network_best_val_t1.pth",
config_dict['pretrained_network_path'])
print("Downloading done.")
config_dict['label_types_test'] += ['pose_mean', 'pose_std']
config_dict['label_types_train'] += ['pose_mean', 'pose_std']
config_dict['latent_dropout'] = 0
config_dict['shuffle_fg'] = False
config_dict['shuffle_3d'] = False
config_dict['actor_subset'] = [1]
#config_dict['useCamBatches'] = 0
from utils import io as utils_io
import os
if 0:
config_dict = utils_io.loadModule(
"./configs/config_train_detect_encode_decode.py").config_dict
network_name = '/network_NVS_best.pth'
else:
config_dict = utils_io.loadModule(
"./configs/config_train_detect_encode_decode_pose.py").config_dict
network_name = '/network_pose_best.pth'
config_dict['num_workers'] = 0
config_dict['label_types_test'].remove('img')
config_dict['label_types_train'].remove('img')
config_dict['batch_size_train'] = 4
config_dict['batch_size_test'] = 4
if 0:
network_path = '../output/train_detectNVS_resL3_layers4_wRGB1_wGrad0_wImgNet2_fg128_ldrop0o3_billinupper_fscale2_shuffleFGFalse_shuffle3dTrue_nth1_cFalse_subNone_bs2_lr0o001_'
config_dict[
'pretrained_network_path'] = network_path + '/models/network_best_val_t1.pth'
else:
network_path = '../examples'
config_dict['pretrained_network_path'] = network_path + network_name
if not os.path.exists(config_dict['pretrained_network_path']):
import urllib.request
print("Downloading pre-trained weights, can take a while...")
urllib.request.urlretrieve(
"https://datasets-cvlab.epfl.ch/RhodinCVPR2019/" + network_name,
config_dict['pretrained_network_path'])
from utils import io as utils_io
import os
# FIXME: get this from the config
network_path = '/bigdata/dingding/cpsc-533/project/UnsupervisedGeometryAwareRepresentationLearning/output/trainNVS_resL3_ResNet_layers4_implRFalse_w3Dp0_w3D0_wRGB1_wGrad0_wImgNet2_skipBG0_fg128_3dTrue_lh3Dp2_ldrop0o3_billinupper_fscale2_shuffleFGTrue_shuffle3dTrue_h36m_nth100_cFalse_sub[1]_bs2_lr0o001_digitCaps10_caps_channel160_capMaskFalse'
config_dict = utils_io.loadModule(
os.path.join(network_path, "config_train_encodeDecode.py")).config_dict
config_dict['pretrained_network_path'] = os.path.join(
network_path, 'models/network_last_val.pth')
#network_path = '/cvlabdata1/home/rhodin/code/humanposeannotation/python/pytorch_selfsupervised_multiview/output/encode_resL3_ResNet_layers4_implRFalse_s3Dp[1_9_11]_w3Dp0_w3D0_wRGB1_wGrad0_wImgNet2_skipBG0_fg128_3dTrue_lh3Dp2_ldrop0o3_billinupper_fscale2_shuffleFGTrue_shuffle3dTrue_h36m_cross_nth1_cFalse_sub[1_5_6_7_8]_bs2_lr0o001_'
# if 0:
# network_path = '../output/trainNVS_resL3_ResNet_layers4_implRFalse_s3Dp[1_9_11]_w3Dp0_w3D0_wRGB1_wGrad0_wImgNet2_skipBG0_fg128_3dTrue_lh3Dp2_ldrop0o3_billinupper_fscale2_shuffleFGTrue_shuffle3dTrue_h36m_cross_nth1_cFalse_sub[1_5_6_7_8]_bs2_lr0o001_'
# config_dict = utils_io.loadModule(network_path + "/config_train_encodeDecode.py").config_dict
# config_dict['pretrained_network_path'] = network_path + '../models/network_last_val.pth'
# else:
# import os
# config_dict = utils_io.loadModule("configs/config_train_encodeDecode.py").config_dict
# config_dict['pretrained_network_path'] = '../examples/network_best_val_t1.pth'
# if not os.path.exists(config_dict['pretrained_network_path']):
# import urllib.request
# print("Downloading pre-trained weights, can take a while...")
# urllib.request.urlretrieve("http://documents.epfl.ch/groups/c/cv/cvlab-unit/www/data/ECCV2018Rhodin/network_best_val_t1.pth",
# config_dict['pretrained_network_path'])
# print("Downloading done.")
config_dict['label_types_test'] += ['pose_mean', 'pose_std']
config_dict['label_types_train'] += ['pose_mean', 'pose_std']
config_dict['latent_dropout'] = 0
config_dict['shuffle_fg'] = False
losses_train = []
losses_test = []
if 'img_crop' in config_dict['output_types']:
if config_dict['loss_weight_rgb']>0:
losses_train.append(image_pixel_loss)
losses_test.append(image_pixel_loss)
if config_dict['loss_weight_imageNet']>0:
losses_train.append(image_imgNet_loss)
losses_test.append(image_imgNet_loss)
loss_train = losses_generic.PreApplyCriterionListDict(losses_train, sum_losses=True)
loss_test = losses_generic.PreApplyCriterionListDict(losses_test, sum_losses=True)
# annotation and pred is organized as a list, to facilitate multiple output types (e.g. heatmap and 3d loss)
return loss_train, loss_test
def get_parameter_description(self, config_dict):#, config_dict):
folder = "../output/trainNVS_{note}_{encoderType}_layers{num_encoding_layers}_implR{implicit_rotation}_w3Dp{loss_weight_pose3D}_w3D{loss_weight_3d}_wRGB{loss_weight_rgb}_wGrad{loss_weight_gradient}_wImgNet{loss_weight_imageNet}_skipBG{latent_bg}_fg{latent_fg}_3d{skip_background}_lh3Dp{n_hidden_to3Dpose}_ldrop{latent_dropout}_billin{upsampling_bilinear}_fscale{feature_scale}_shuffleFG{shuffle_fg}_shuffle3d{shuffle_3d}_{training_set}_nth{every_nth_frame}_c{active_cameras}_sub{actor_subset}_bs{useCamBatches}_lr{learning_rate}_digitCaps{num_digit_caps}_caps_channel{num_caps_out_channel}_capMask{caps_masked}".format(**config_dict)
folder = folder.replace(' ','').replace('../','[DOT_SHLASH]').replace('.','o').replace('[DOT_SHLASH]','../').replace(',','_')
#config_dict['storage_folder'] = folder
return folder
if __name__ == "__main__":
config_dict_module = utils_io.loadModule("config_train_encodeDecode.py")
config_dict = config_dict_module.config_dict
ignite = IgniteTrainNVS()
ignite.run(config_dict_module.__file__, config_dict)
