import matplotlib.pyplot as plt
import easydict
import os
FLAGS = easydict.EasyDict({
"img_size": 256,
"load_size": 266,
"batch_size": 1,
"epochs": 200,
"lr": 0.0002,
"n_classes": 24,
"in_txt_path":
"D:/[1]DB/[2]third_paper_DB/[4]Age_and_gender/race_age_gender_generation/Morph_AFAD_16_63/first_fold/AFAD-M_Morph-F_16_39_40_63/train/male_16_39_train.txt",
"in_img_path":
"D:/[1]DB/[1]second_paper_DB/AFAD_16_69_DB/backup/fix_AFAD/",
"st_txt_path":
"D:/[1]DB/[2]third_paper_DB/[4]Age_and_gender/race_age_gender_generation/Morph_AFAD_16_63/first_fold/AFAD-M_Morph-F_16_39_40_63/train/female_40_63_train.txt",
"st_img_path":
"D:/[1]DB/[2]third_paper_DB/[4]Age_and_gender/Morph/All/female_40_63/",
"train": True,
"pre_checkpoint": False,
"pre_checkpoint_path": "",
"save_checkpoint": "",
"save_images": "C:/Users/Yuhwan/Pictures/test_sample",
"graphs": ""
})
g_optim = tf.keras.optimizers.Adam(FLAGS.lr, beta_1=0.5, beta_2=0.5)
d_optim = tf.keras.optimizers.Adam(FLAGS.lr, beta_1=0.5, beta_2=0.5)
s_optim = tf.keras.optimizers.Adam(FLAGS.lr, beta_1=0.5, beta_2=0.5)
import past_models
import dataset
from main_dcgan import train
import easydict
opt = easydict.EasyDict({
'n_epochs': 200,
'batch_size': 128,
'lr': 0.0002,
'b1': 0.5,
'b2': 0.999,
'n_cpu': 8,
'latent_dim': 128,
'img_size': 32,
'n_critic': 1,
'clip_value': 0.01,
'sample_interval': 100,
'log_interval': 10,
'dataset': 'mnist',
'num_filters': 128, #for CNN Discriminator and Generator
'saveDir': None,
'resume': None,
'loadDir': None
})
_, dataloader = dataset.makeDataloader(opt)
# Initialize generator and discriminator
generator = past_models.DCGANGenerator32(opt)
discriminator = past_models.DCGANDiscriminator32(opt)
import torch.optim as optim
from torch.distributions import Normal
import matplotlib.pyplot as plt
from quad_env import QuadEnv
# Cart Pole
# based on:
# https://github.com/pytorch/examples/blob/master/reinforcement_learning/actor_critic.py
# args = parser.parse_args()
args = easydict.EasyDict({
"gamma": 0.99,
"seed": 203,
"render": False,
"log_interval": 10,
"write_logger": True
})
# env = gym.make('LunarLanderContinuous-v2')
env = QuadEnv()
# env.seed(args.seed)
torch.manual_seed(args.seed)
SavedAction = namedtuple('SavedAction', ['log_prob', 'value'])
# state_dim = env.observation_space.shape[0]
state_dim = 21
# action_dim = env.action_space.shape[0]
action_dim = 4
def _forward_test(self, input):
cnn_features = input
arg = easydict.EasyDict({
'clip_boxes': self.test_clip_boxes,
'nms_thresh': self.test_nms_thresh,
'max_proposals': self.test_max_proposals
})
# Make sure that setImageSize has been called
assert self.image_height and self.image_width and not self._called_forward_size, \
'Must call setImageSize before each forward pass'
self._called_forward_size = True
rpn_out, act_reg = self.rpn.forward(cnn_features)
rpn_boxes, rpn_anchors, rpn_trans, rpn_scores = rpn_out
num_boxes = rpn_boxes.size(1)
# Maybe clip boxes to image boundary
if arg.clip_boxes:
bounds = {
'x_min': 1,
'y_min': 1,
'x_max': self.image_width,
'y_max': self.image_height
}
rpn_boxes, valid = box_utils.clip_boxes(rpn_boxes, bounds,
'xcycwh')
#print(string.format('%d/%d boxes are predicted valid',
# torch.sum(valid), valid:nElement()))
#Clamp parallel arrays only to valid boxes (not oob of the image)
rpn_boxes = self.clamp_data(rpn_boxes, valid)
rpn_anchors = self.clamp_data(rpn_anchors, valid)
rpn_trans = self.clamp_data(rpn_trans, valid)
rpn_scores = self.clamp_data(rpn_scores, valid)
num_boxes = rpn_boxes.size(1)
# Convert rpn boxes from (xc, yc, w, h) format to (x1, y1, x2, y2)
rpn_boxes_x1y1x2y2 = box_utils.xcycwh_to_x1y1x2y2(rpn_boxes[0])
# Convert objectness positive / negative scores to probabilities
rpn_scores_exp = torch.exp(rpn_scores)
pos_exp = rpn_scores_exp[0, :, 0]
neg_exp = rpn_scores_exp[0, :, 1]
scores = (pos_exp + neg_exp).pow(-1) * pos_exp
verbose = False
if verbose:
print('in LocalizationLayer forward_test')
print('Before NMS there are %d boxes' % num_boxes)
print('Using NMS threshold %f' % arg.nms_thresh)
#Run NMS and sort by objectness score
boxes_scores = torch.cat((rpn_boxes_x1y1x2y2, scores.view(-1, 1)),
dim=1)
if arg.max_proposals == -1:
idx = box_utils.nms(boxes_scores.data, arg.nms_thresh)
else:
idx = box_utils.nms(boxes_scores.data, arg.nms_thresh,
arg.max_proposals)
rpn_boxes_nms = torch.squeeze(rpn_boxes)[idx]
if verbose:
print('After NMS there are %d boxes' % rpn_boxes_nms.size(0))
output = rpn_boxes_nms
return output
FLAGS = easydict.EasyDict({"img_height": 128,
"img_width": 88,
"tr_txt_path": "D:/[1]DB/[4]etc_experiment/Body_age/OULP-Age/train.txt",
"tr_txt_name": "D:/[1]DB/[4]etc_experiment/Body_age/OULP-Age/GEI_IDList_train.txt",
"tr_img_path": "D:/[1]DB/[4]etc_experiment/Body_age/OULP-Age/GEI/",
"te_txt_path": "D:/[1]DB/[4]etc_experiment/Body_age/OULP-Age/test.txt",
"te_txt_name": "D:/[1]DB/[4]etc_experiment/Body_age/OULP-Age/GEI_IDList_test_fix.txt",
"te_img_path": "D:/[1]DB/[4]etc_experiment/Body_age/OULP-Age/GEI/",
"batch_size": 200,
"epochs": 300,
"num_classes": 86,
"lr": 0.0001,
"save_checkpoint": "",
"graphs": "C:/Users/Yuhwan/Downloads/",
"train": True,
"pre_checkpoint": False,
"pre_checkpoint_path": ""})
import wget
import easydict
import matplotlib.pyplot as plt
import time
import numpy as np
import cv2
import face_embedding
import argparse
import os
import logging
parser = argparse.ArgumentParser(description='face model test')
args = easydict.EasyDict()
args["image_size"] = '112,112'
args["model"] = '../models/model,0'
args["gpu"] = "-1"
args["det"] = 2
args["flip"] = 0
args["threshold"] = 1.24
model = face_embedding.FaceModel(args)
people_list = ["victor", "zhang", "zhou"]
info = [
"You have a dental appointment today at 10:30 am in Santa Clara Hospital",
"Your daughter will visit you at 12 am at your home.",
"Please don't forget to take pills every 8 hours from 8 am."
]
print("Here are the people we recognize: {}".format(people_list))
imgs_list = [[cv2.imread("{}/{}.jpeg".format(person, i)) for i in range(1, 5)]
for person in people_list]
physical_devices = tf.config.experimental.list_physical_devices('GPU')
if len(physical_devices) > 0:
tf.config.experimental.set_memory_growth(physical_devices[0], True)
from absl.flags import FLAGS
from tensorflow.python.saved_model import tag_constants
import cv2
import numpy as np
import easydict
FLAGS = easydict.EasyDict(
{
'framework' : 'tf',
'weights' : './checkpoints/yolov4-tiny-416',
'size' : 416,
'tiny' : True,
'model' : 'yolov4',
'iou' : 0.30,
'score' : 0.30
}
)
class PeopleDetector:
def __init__(self):
self.input_size = FLAGS.size
self.saved_model_loaded = tf.saved_model.load(FLAGS.weights, tags=[tag_constants.SERVING])
self.infer = self.saved_model_loaded.signatures['serving_default']
def detect(self, frame) -> (bool, list, list) :
input_size = self.input_size
infer = self.infer
from random import *
model_names = sorted(name for name in models.__dict__
if name.islower() and not name.startswith("__")
and callable(models.__dict__[name]))
args = easydict.EasyDict({
"arch": 'resnet18',
"root": "/home/diml/ddrive/dataset/deepfake_1st",
"train_list": "train_list_1st.txt",
"test_list": "test_list_1st.txt",
"epochs": 200,
"batch_size": 2,
"lr": 0.01,
"momentum": 0.9,
"weight_decay": 5e-4,
"print_freq": 100,
"eval_freq": 10,
"workers": 4,
#"resume":"10fgsm_b7_checkpoint.pth.tar",
"resume": False,
"pretrained": False,
"evaluate": False,
"start_epoch": 0,
"gpu": 0,
})
best_acc1 = 0
# for fgsm image generation
def fgsm_attack(model, loss, images, labels, eps):
def __init__(self, dataset, split, lr, save_path, l1_coef, l2_coef,
pre_trained_gen, pre_trained_disc, val_pre_trained_gen,
val_pre_trained_disc, batch_size, num_workers, epochs,
dataset_paths, arrangement, sampling):
with open('config.yaml',
'r') as f: # Wsteczna kompatybilnosc dla Text2ImageDataset
config = yaml.safe_load(f)
self.generator = torch.nn.DataParallel(
gan_factory.generator_factory('gan').cuda())
self.discriminator = torch.nn.DataParallel(
gan_factory.discriminator_factory('gan').cuda())
if pre_trained_disc:
self.discriminator.load_state_dict(torch.load(pre_trained_disc))
else:
self.discriminator.apply(Utils.weights_init)
if pre_trained_gen:
self.generator.load_state_dict(torch.load(pre_trained_gen))
else:
self.generator.apply(Utils.weights_init)
self.val_pre_trained_gen = val_pre_trained_gen
self.val_pre_trained_disc = val_pre_trained_disc
self.arrangement = arrangement
self.sampling = sampling
if dataset == 'birds': # Wsteczna kompatybilnosc dla Text2ImageDataset
self.dataset = Text2ImageDataset(
config['birds_dataset_path'], split=split
) # '...\Text2Image\datasets\ee285f-public\caltech_ucsd_birds\birds.hdf5'
elif dataset == 'flowers': # Wsteczna kompatybilnosc dla Text2ImageDataset
self.dataset = Text2ImageDataset(
config['flowers_dataset_path'], split=split
) # '...\Text2Image\datasets\ee285f-public\oxford_flowers\flowers.hdf5'
elif dataset == 'live':
self.dataset_dict = easydict.EasyDict(dataset_paths)
self.dataset = Text2ImageDataset2(
datasetFile=self.dataset_dict.datasetFile,
imagesDir=self.dataset_dict.imagesDir,
textDir=self.dataset_dict.textDir,
split=split,
arrangement=arrangement,
sampling=sampling)
else:
print('Dataset not supported.')
print('Images =', len(self.dataset))
self.noise_dim = 100
self.batch_size = batch_size
self.num_workers = num_workers
self.lr = lr
self.beta1 = 0.5
self.num_epochs = epochs
self.l1_coef = l1_coef
self.l2_coef = l2_coef
self.data_loader = DataLoader(
self.dataset,
batch_size=self.batch_size,
shuffle=True,
num_workers=self.num_workers
) # shuffle=True - przetasowuje zbior danych w kazdej epoce
self.optimD = torch.optim.Adam(self.discriminator.parameters(),
lr=self.lr,
betas=(self.beta1, 0.999))
self.optimG = torch.optim.Adam(self.generator.parameters(),
lr=self.lr,
betas=(self.beta1, 0.999))
self.checkpoints_path = 'checkpoints'
self.save_path = save_path
self.loss_filename_t = 'gd_loss_t.csv'
self.loss_filename_v = 'gd_loss_v.csv'
def testTrainableVariables(self):
inputs = tf.placeholder(tf.float32, [1, 224, 224, 3])
model = TruncatedBaseNetwork(
easydict.EasyDict({
'architecture': 'resnet_v1_50',
'endpoint': 'block4/unit_3/bottleneck_v1/conv2',
'freeze_tail': False,
'use_tail': True,
})
)
model(inputs)
# Variables in ResNet-50:
# (the order of beta and gamma depends on the TensorFlow's version)
# 0 conv1/weights:0
# 1 conv1/BatchNorm/(beta|gamma):0
# 2 conv1/BatchNorm/(beta|gamma):0
# 3 block1/unit_1/bottleneck_v1/shortcut/weights:0
# (...)
# 153 block4/unit_3/bottleneck_v1/conv2/weights:0
# 154 block4/unit_3/bottleneck_v1/conv2/BatchNorm/(beta|gamma):0
# 155 block4/unit_3/bottleneck_v1/conv2/BatchNorm/(beta|gamma):0
# --- endpoint ---
# 156 block4/unit_3/bottleneck_v1/conv3/weights:0
# 157 block4/unit_3/bottleneck_v1/conv3/BatchNorm/(beta|gamma):0
# 158 block4/unit_3/bottleneck_v1/conv3/BatchNorm/(beta|gamma):0
# 159 logits/weights:0
# 160 logits/biases:0
trainable_vars = model.get_trainable_vars()
self.assertEqual(len(trainable_vars), 156)
self.assertEqual(
trainable_vars[-3].name,
'truncated_base_network/resnet_v1_50/' +
'block4/unit_3/bottleneck_v1/conv2/weights:0'
)
model = TruncatedBaseNetwork(
easydict.EasyDict({
'architecture': 'resnet_v1_50',
'endpoint': 'block4/unit_2/bottleneck_v1/conv2',
'fine_tune_from': 'block4/unit_2/bottleneck_v1/conv1',
'freeze_tail': False,
'use_tail': True,
})
)
model(inputs)
trainable_vars = model.get_trainable_vars()
# Now there should be only 6 trainable vars:
# 141 block4/unit_2/bottleneck_v1/conv1/weights:0
# 142 block4/unit_2/bottleneck_v1/conv1/BatchNorm/beta:0
# 143 block4/unit_2/bottleneck_v1/conv1/BatchNorm/gamma:0
# 144 block4/unit_2/bottleneck_v1/conv2/weights:0
# 145 block4/unit_2/bottleneck_v1/conv2/BatchNorm/beta:0
# 146 block4/unit_2/bottleneck_v1/conv2/BatchNorm/gamma:0
self.assertEqual(len(trainable_vars), 6)
#
# Check that we return no vars if fine_tune_from is after the chosen
# endpoint (there is nothing to fine-tune!) and tail is frozen.
#
model = TruncatedBaseNetwork(
easydict.EasyDict(
{
'architecture': 'resnet_v1_50',
'endpoint': 'block4/unit_2/bottleneck_v1/conv1',
'fine_tune_from': 'block4/unit_2/bottleneck_v1/conv2',
'freeze_tail': True,
'use_tail': True,
}
)
)
model(inputs)
self.assertEqual(len(model.get_trainable_vars()), 0)
elif options.mode == 1:
model.sample()
options = easydict.EasyDict({
'batch': 100,
'seed': random.randint(0, 2**31 - 1),
'name': 'CGAN',
'mode': 1,
'dataset': 'linnaeus5',
'dataset_path': './gdrive/My Drive/dataset',
'checkpoints_path': './gdrive/My Drive/dataset/checkpoints',
'batch_size': 30,
'epochs': 201,
'lr': 3e-4,
'lr_decay': True,
'lr_decay_rate': 0.1,
'lr_decay_steps': 5e5,
'gpu_ids': '0',
'save': True,
'save_interval': 1200 / 30,
'sample': True,
'sample_size': 10,
'sample_interval': 1200 / 30,
'training': 0
})
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
options.training = options.mode != 1
options.dataset_path += ('/' + options.dataset)
from torch.utils.tensorboard import SummaryWriter
from tqdm import tqdm, tqdm_notebook, trange
from configs import getConfig
from model.data_loader import kfold_data_generator
from model.dataset import MSI_train_Dataset, MSI_valid_Dataset
from model.net import fpn, unet
from utils import save_setting_info
# arg = getConfig()
arg = easydict.EasyDict({
"model_name": 'unet',
"k_folds": 1,
"BACKBONE": 'resnet34',
"PRETRAINED": None,
"epochs": 300,
"BATCH_SIZE": 32,
'lr': 0.001,
'WORKERS': 0,
'Threshold': 0.7
})
# 구글스프레드시트 설정
scope = [
'https://spreadsheets.google.com/feeds',
'https://www.googleapis.com/auth/drive',
]
json_file_name = 'money-check-260910-c217b2d4ba6a.json'
credentials = ServiceAccountCredentials.from_json_keyfile_name(
import torch.optim as optim
import torch.utils.data as data
import torchvision
import torch.nn.functional as F
from torchvision import models, transforms
# -
torch.manual_seed(1234)
np.random.seed(1234)
random.seed(1234)
args = easydict.EasyDict({
"train_root_dir": "../data/main/train",
"val_root_dir": "../data/main/val",
"batch_size": 4,
"size": (480, 640),
"num_worker": 1,
"lr": 1e-4,
"start_epoch": 1,
"result_path": './results_AODNet',
})
device_str = "cuda:3"
device = torch.device(device_str if torch.cuda.is_available() else "cpu")
class HazeDataset(data.Dataset):
def __init__(self, root_dir, size=(480, 640)):
self.root_dir, _, self.files = next(os.walk(root_dir))
self.size = size
def __len__(self):
def main():
args = easydict.EasyDict({
"dataroot": "/mnt/gold/users/s18150/mywork/pytorch/data/gan",
"save_dir": "./",
"prefix": "minibach_feature",
"workers": 8,
"batch_size": 128,
"image_size": 64,
"nc": 3,
"nz": 100,
"ngf": 64,
"ndf": 64,
"epochs": 50,
"lr": 0.0002,
"beta1": 0.5,
"gpu": 9,
"use_cuda": True,
"feature_matching": True,
"mini_batch": True
})
manualSeed = 999
random.seed(manualSeed)
torch.manual_seed(manualSeed)
device = torch.device(
'cuda:{}'.format(args.gpu) if args.use_cuda else 'cpu')
transform = transforms.Compose([
transforms.Resize(args.image_size),
transforms.CenterCrop(args.image_size),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
dataset = dset.ImageFolder(root=args.dataroot, transform=transform)
dataloader = torch.utils.data.DataLoader(dataset,
batch_size=args.batch_size,
shuffle=True,
num_workers=args.workers)
# Generator用のモデルのインスタンス作成
netG = net.Generator(args.nz, args.ngf, args.nc).to(device)
# Generator用のモデルの初期値を設定
netG.apply(net.weights_init)
# Discriminator用のモデルのインスタンス作成
netD = net.Discriminator(args.nc, args.ndf, device, args.batch_size,
args.mini_batch).to(device)
# Discriminator用のモデルの初期値を設定
netD.apply(net.weights_init)
# BCE Loss classのインスタンスを作成
criterionD = nn.BCELoss()
if args.feature_matching is True:
criterionG = nn.MSELoss(reduction='elementwise_mean')
else:
criterionG = nn.BCELoss()
# Generatorに入力するノイズをバッチごとに作成 (バッチ数は64)
# これはGeneratorの結果を描画するために使用する
fixed_noise = torch.randn(64, args.nz, 1, 1, device=device)
# 最適化関数のインスタンスを作成
optimizerD = optim.Adam(netD.parameters(),
lr=args.lr,
betas=(args.beta1, 0.999))
optimizerG = optim.Adam(netG.parameters(),
lr=args.lr,
betas=(args.beta1, 0.999))
r = run.NNRun(netD, netG, optimizerD, optimizerG, criterionD, criterionG,
device, fixed_noise, args)
# 学習
r.train(dataloader)
def GlowCS(args):
if args.init_norms == None:
args.init_norms = [None] * len(args.m)
else:
assert args.init_strategy == "random_fixed_norm", "init_strategy should be random_fixed_norm if init_norms is used"
assert len(args.m) == len(args.gamma) == len(
args.init_norms
), "length of either m, gamma or init_norms are not same"
loopOver = zip(args.m, args.gamma, args.init_norms)
for m, gamma, init_norm in loopOver:
skip_to_next = False # flag to skip to next loop if recovery is fails due to instability
n = args.size * args.size * 3
modeldir = "./trained_models/%s/glow" % args.model
test_folder = "./test_images/%s" % args.dataset
save_path = "./results/%s/%s" % (args.dataset, args.experiment)
# loading dataset
trans = transforms.Compose(
[transforms.Resize((args.size, args.size)),
transforms.ToTensor()])
test_dataset = datasets.ImageFolder(test_folder, transform=trans)
test_dataloader = torch.utils.data.DataLoader(
test_dataset,
batch_size=args.batchsize,
drop_last=False,
shuffle=False)
# loading glow configurations
config_path = modeldir + "/configs.json"
with open(config_path, 'r') as f:
configs = json.load(f)
# sensing matrix
A = np.random.normal(0, 1 / np.sqrt(m), size=(n, m))
A = torch.tensor(A,
dtype=torch.float,
requires_grad=False,
device=args.device)
# regularizor
gamma = torch.tensor(gamma,
requires_grad=True,
dtype=torch.float,
device=args.device)
# adding noise
if args.noise == "random_bora":
noise = np.random.normal(0, 1, size=(args.batchsize, m))
noise = noise * 0.1 / np.sqrt(m)
noise = torch.tensor(noise,
dtype=torch.float,
requires_grad=False,
device=args.device)
else:
noise = np.random.normal(0, 1, size=(args.batchsize, m))
noise = noise / (np.linalg.norm(noise, 2, axis=-1,
keepdims=True)) * float(args.noise)
noise = torch.tensor(noise,
dtype=torch.float,
requires_grad=False,
device=args.device)
# start solving over batches
Original = []
Recovered = []
Z_Recovered = []
Residual_Curve = []
Recorded_Z = []
for i, data in enumerate(test_dataloader):
x_test = data[0]
x_test = x_test.clone().to(device=args.device)
n_test = x_test.size()[0]
assert n_test == args.batchsize, "please make sure that no. of images are evenly divided by batchsize"
# loading glow model
glow = Glow((3, args.size, args.size),
K=configs["K"],
L=configs["L"],
coupling=configs["coupling"],
n_bits_x=configs["n_bits_x"],
nn_init_last_zeros=configs["last_zeros"],
device=args.device)
glow.load_state_dict(torch.load(modeldir + "/glowmodel.pt"))
glow.eval()
# making a forward to record shapes of z's for reverse pass
_ = glow(glow.preprocess(torch.zeros_like(x_test)))
# initializing z from Gaussian with std equal to init_std
if args.init_strategy == "random":
z_sampled = np.random.normal(0, args.init_std, [n_test, n])
z_sampled = torch.tensor(z_sampled,
requires_grad=True,
dtype=torch.float,
device=args.device)
# intializing z from Gaussian and scaling its norm to init_norm
elif args.init_strategy == "random_fixed_norm":
z_sampled = np.random.normal(0, 1, [n_test, n])
z_sampled = z_sampled / np.linalg.norm(
z_sampled, axis=-1, keepdims=True)
z_sampled = z_sampled * init_norm
z_sampled = torch.tensor(z_sampled,
requires_grad=True,
dtype=torch.float,
device=args.device)
print("z intialized with a norm equal to = %0.1f" % init_norm)
# initializing z from pseudo inverse
elif args.init_strategy == "pseudo_inverse":
x_test_flat = x_test.view([-1, n])
y_true = torch.matmul(x_test_flat, A) + noise
A_pinv = torch.pinverse(A)
x_pinv = torch.matmul(y_true, A_pinv)
x_pinv = x_pinv.view([-1, 3, args.size, args.size])
x_pinv = torch.clamp(x_pinv, 0, 1)
z, _, _ = glow(glow.preprocess(x_pinv * 255, clone=True))
z = glow.flatten_z(z).clone().detach()
z_sampled = torch.tensor(z,
requires_grad=True,
dtype=torch.float,
device=args.device)
# initializing z from a solution of lasso-wavelet
elif args.init_strategy == "lasso_wavelet":
new_args = {
"batch_size": n_test,
"lmbd": 0.01,
"lasso_solver": "sklearn"
}
new_args = easydict.EasyDict(new_args)
estimator = celebA_estimators.lasso_wavelet_estimator(new_args)
x_ch_last = x_test.permute(0, 2, 3, 1)
x_ch_last = x_ch_last.contiguous().view([-1, n])
y_true = torch.matmul(x_ch_last, A) + noise
x_lasso = estimator(
np.sqrt(2 * m) * A.data.cpu().numpy(),
np.sqrt(2 * m) * y_true.data.cpu().numpy(), new_args)
x_lasso = np.array(x_lasso)
x_lasso = x_lasso.reshape(-1, 64, 64, 3)
x_lasso = x_lasso.transpose(0, 3, 1, 2)
x_lasso = torch.tensor(x_lasso,
dtype=torch.float,
device=args.device)
z, _, _ = glow(x_lasso - 0.5)
z = glow.flatten_z(z).clone().detach()
z_sampled = torch.tensor(z,
requires_grad=True,
dtype=torch.float,
device=args.device)
print("z intialized from a solution of lasso-wavelet")
# intializing z from null(A)
elif args.init_strategy == "null_space":
x_test_flat = x_test.view([-1, n])
x_test_flat_np = x_test_flat.data.cpu().numpy()
A_np = A.data.cpu().numpy()
nullA = null_space(A_np.T)
coeff = np.random.normal(0, 1,
(args.batchsize, nullA.shape[1]))
x_null = np.array([(nullA * c).sum(axis=-1) for c in coeff])
pert_norm = 5 # <-- 5 gives optimal results -- bad initialization and not too unstable
x_null = x_null / np.linalg.norm(x_null, axis=1,
keepdims=True) * pert_norm
x_perturbed = x_test_flat_np + x_null
# no clipping x_perturbed to make sure forward model is ||y-Ax|| is the same
err = np.matmul(x_test_flat_np, A_np) - np.matmul(
x_perturbed, A_np)
assert (err**2).sum(
) < 1e-6, "null space does not satisfy ||y-A(x+x0)|| <= 1e-6"
x_perturbed = x_perturbed.reshape(-1, 3, args.size, args.size)
x_perturbed = torch.tensor(x_perturbed,
dtype=torch.float,
device=args.device)
z, _, _ = glow(x_perturbed - 0.5)
z = glow.flatten_z(z).clone().detach()
z_sampled = torch.tensor(z,
requires_grad=True,
dtype=torch.float,
device=args.device)
print("z initialized from a point in null space of A")
else:
raise "Initialization strategy not defined"
# selecting optimizer
if args.optim == "adam":
optimizer = torch.optim.Adam(
[z_sampled],
lr=args.lr,
)
elif args.optim == "lbfgs":
optimizer = torch.optim.LBFGS(
[z_sampled],
lr=args.lr,
)
else:
raise "optimizer not defined"
# to be recorded over iteration
psnr_t = torch.nn.MSELoss().to(device=args.device)
residual = []
recorded_z = []
# running optimizer steps
for t in range(args.steps):
def closure():
optimizer.zero_grad()
z_unflat = glow.unflatten_z(z_sampled, clone=False)
x_gen = glow(z_unflat, reverse=True, reverse_clone=False)
x_gen = glow.postprocess(x_gen, floor_clamp=False)
x_test_flat = x_test.view([-1, n])
x_gen_flat = x_gen.view([-1, n])
y_true = torch.matmul(x_test_flat, A) + noise
y_gen = torch.matmul(x_gen_flat, A)
global residual_t
residual_t = ((y_gen - y_true)**2).sum(dim=1).mean()
z_reg_loss_t = gamma * z_sampled.norm(dim=1).mean()
loss_t = residual_t + z_reg_loss_t
psnr = psnr_t(x_test, x_gen)
psnr = 10 * np.log10(1 / psnr.item())
print(
"\rAt step=%0.3d|loss=%0.4f|residual=%0.4f|z_reg=%0.5f|psnr=%0.3f"
% (t, loss_t.item(), residual_t.item(),
z_reg_loss_t.item(), psnr),
end="\r")
loss_t.backward()
return loss_t
try:
optimizer.step(closure)
recorded_z.append(z_sampled.data.cpu().numpy())
residual.append(residual_t.item())
except:
# try may not work due to instability in the reverse direction.
skip_to_next = True
break
if skip_to_next:
break
# getting recovered and true images
with torch.no_grad():
x_test_np = x_test.data.cpu().numpy().transpose(0, 2, 3, 1)
z_unflat = glow.unflatten_z(z_sampled, clone=False)
x_gen = glow(z_unflat, reverse=True, reverse_clone=False)
x_gen = glow.postprocess(x_gen, floor_clamp=False)
x_gen_np = x_gen.data.cpu().numpy().transpose(0, 2, 3, 1)
x_gen_np = np.clip(x_gen_np, 0, 1)
z_recov = z_sampled.data.cpu().numpy()
Original.append(x_test_np)
Recovered.append(x_gen_np)
Z_Recovered.append(z_recov)
Residual_Curve.append(residual)
Recorded_Z.append(recorded_z)
# freeing up memory for second loop
glow.zero_grad()
optimizer.zero_grad()
del x_test, x_gen, optimizer, psnr_t, z_sampled, glow
torch.cuda.empty_cache()
print("\nbatch completed")
if skip_to_next:
print(
"\nskipping current loop due to instability or user triggered quit"
)
continue
# collecting everything together
Original = np.vstack(Original)
Recovered = np.vstack(Recovered)
Z_Recovered = np.vstack(Z_Recovered)
Recorded_Z = np.vstack(Recorded_Z)
psnr = [compare_psnr(x, y) for x, y in zip(Original, Recovered)]
z_recov_norm = np.linalg.norm(Z_Recovered, axis=-1)
# print performance analysis
printout = "+-" * 10 + "%s" % args.dataset + "-+" * 10 + "\n"
printout = printout + "\t n_test = %d\n" % len(Recovered)
printout = printout + "\t n = %d\n" % n
printout = printout + "\t m = %d\n" % m
printout = printout + "\t gamma = %0.6f\n" % gamma
printout = printout + "\t optimizer = %s\n" % args.optim
printout = printout + "\t lr = %0.3f\n" % args.lr
printout = printout + "\t steps = %0.3f\n" % args.steps
printout = printout + "\t init_strategy = %s\n" % args.init_strategy
printout = printout + "\t init_std = %0.3f\n" % args.init_std
if init_norm is not None:
printout = printout + "\t init_norm = %0.3f\n" % init_norm
printout = printout + "\t z_recov_norm = %0.3f\n" % np.mean(
z_recov_norm)
printout = printout + "\t PSNR = %0.3f\n" % (np.mean(psnr))
print(printout)
# saving printout
if args.save_metrics_text:
with open("%s_cs_glow_results.txt" % args.dataset, "a") as f:
f.write('\n' + printout)
# setting folder to save results in
if args.save_results:
gamma = gamma.item()
file_names = [
name[0].split("/")[-1] for name in test_dataset.samples
]
if args.init_strategy == "random":
save_path_template = save_path + "/cs_m_%d_gamma_%0.6f_steps_%d_lr_%0.3f_init_std_%0.2f_optim_%s"
save_path = save_path_template % (
m, gamma, args.steps, args.lr, args.init_std, args.optim)
elif args.init_strategy == "random_fixed_norm":
save_path_template = save_path + "/cs_m_%d_gamma_%0.6f_steps_%d_lr_%0.3f_init_%s_%0.3f_optim_%s"
save_path = save_path_template % (m, gamma, args.steps,
args.lr, args.init_strategy,
init_norm, args.optim)
else:
save_path_template = save_path + "/cs_m_%d_gamma_%0.6f_steps_%d_lr_%0.3f_init_%s_optim_%s"
save_path = save_path_template % (m, gamma, args.steps,
args.lr, args.init_strategy,
args.optim)
if not os.path.exists(save_path):
os.makedirs(save_path)
else:
save_path_1 = save_path + "_1"
if not os.path.exists(save_path_1):
os.makedirs(save_path_1)
save_path = save_path_1
else:
save_path_2 = save_path + "_2"
if not os.path.exists(save_path_2):
os.makedirs(save_path_2)
save_path = save_path_2
# saving results now
_ = [
sio.imsave(save_path + "/" + name, x)
for x, name in zip(Recovered, file_names)
]
Residual_Curve = np.array(Residual_Curve).mean(axis=0)
np.save(save_path + "/original.npy", Original)
np.save(save_path + "/recovered.npy", Recovered)
np.save(save_path + "/z_recovered.npy", Z_Recovered)
np.save(save_path + "/residual_curve.npy", Residual_Curve)
if init_norm is not None:
np.save(save_path + "/Recorded_Z_init_norm_%d.npy" % init_norm,
Recorded_Z)
torch.cuda.empty_cache()
import easydict
import torchtext
import torchtext.data as data
import torchtext.datasets as datasets
import datetime
import pickle
import dill
import re
import timeit
import operator
args = easydict.EasyDict({
"embed_num": 3257,
"embed_dim": 128,
"kernel_num": 100,
"class_num": 3
})
FOOD_QM = "data/food/MEM data/QM.txt"
SHOPPING_QM = "data/shopping/MEM data/QM.txt"
WEATHER_QM = "data/weather/MEM data/QM.txt"
FOOD_POS = "data/food/MEM data/positive_org.txt"
FOOD_NEG = "data/food/MEM data/negative_org.txt"
SHOPPING_POS = "data/shopping/MEM data/positive_org.txt"
SHOPPING_NEG = "data/shopping/MEM data/negative_org.txt"
WEATHER_POS = "data/weather/MEM data/positive_org.txt"
WEATHER_NEG = "data/weather/MEM data/negative_org.txt"
if torch.cuda.is_available():
def DCTCS(args):
loopOver = zip(args.m, args.gamma)
for m, gamma in loopOver:
n = args.size * args.size * 3
test_folder = "./test_images/%s/imgs" % args.dataset
save_path = "./results/%s/%s" % (args.dataset, args.experiment)
divide_by = 255 # "max" or 255 or None
# loading test images
x_test = [PIL.Image.open(p) for p in glob(test_folder + "/*")]
file_names = [name.split("/")[-1] for name in glob(test_folder + "/*")]
x_test = [
img.resize((args.size, args.size), PIL.Image.BILINEAR) if
(img.size[0] != 64) else img for img in x_test
]
x_test = [np.array(img) for img in x_test]
x_test = np.array(x_test)
# normalizing images
if divide_by == "max":
x_test = x_test / x_test.max(axis=(1, 2, 3), keepdims=True)
elif divide_by == 255:
x_test = x_test / 255
elif divide_by == None:
pass
n_test = x_test.shape[0]
# arg parser to pass to solver methods
new_args = {
"batch_size": n_test,
"lmbd": gamma,
"lasso_solver": "sklearn"
}
new_args = easydict.EasyDict(new_args)
estimator = celebA_estimators.lasso_dct_estimator(new_args)
A = np.random.normal(0, 1 / np.sqrt(m), size=(n, m))
# adding noise
if args.noise == "random_bora":
noise = np.random.normal(0, 1, size=(n_test, m))
noise = noise * 0.1 / np.sqrt(m)
else:
noise = np.random.normal(0, 1, size=(n_test, m))
noise = noise / (np.linalg.norm(noise, 2, axis=-1,
keepdims=True)) * float(args.noise)
y_true = np.matmul(x_test.reshape(n_test, -1), A) + noise
x_hat = estimator(
np.sqrt(2 * m) * A,
np.sqrt(2 * m) * y_true, new_args)
x_hat = np.array(x_hat)
x_hat = x_hat.reshape(-1, 64, 64, 3)
x_hat = np.clip(x_hat, 0, 1)
psnr = [compare_psnr(x, xhat) for x, xhat in zip(x_test, x_hat)]
# print performance analysis
printout = "+-" * 10 + "%s" % args.dataset + "-+" * 10 + "\n"
printout = printout + "\t n_test = %d\n" % len(x_hat)
printout = printout + "\t n = %d\n" % n
printout = printout + "\t m = %d\n" % m
printout = printout + "\t solver = lasso_dct\n"
printout = printout + "\t gamma = %0.8f\n" % gamma
printout = printout + "\t PSNR = %0.3f \n" % np.mean(psnr)
print(printout)
if args.save_metrics_text:
with open("%s_cs_dct_results.txt" % args.dataset, "a") as f:
f.write('\n' + printout)
# saving images
if args.save_results:
save_path_template = save_path + "/cs_m_%d_lasso_dct_gamma_%0.8f"
save_path = save_path_template % (m, gamma)
if not os.path.exists(save_path):
os.makedirs(save_path)
else:
save_path_1 = save_path + "_1"
if not os.path.exists(save_path_1):
os.makedirs(save_path_1)
save_path = save_path_1
else:
save_path_2 = save_path + "_2"
if not os.path.exists(save_path_2):
os.makedirs(save_path_2)
save_path = save_path_2
_ = [
sio.imsave(save_path + "/" + name, x)
for x, name in zip(x_hat, file_names)
]
# _ = [sio.imsave(save_path+"/"+name.split(".")[0]+".jpg", x, quality=100) for x,name in zip(x_hat,file_names)]
np.save(save_path + "/original.npy", x_test)
np.save(save_path + "/recovered.npy", x_hat)
args = easydict.EasyDict({
"src_font":
'C:/Users/1-16/python/neural-fonts-master/fonts/NanumGothic.ttf',
"dst_font":
'C:/Users/1-16/python/neural-fonts-master/fonts/NanumGothic.ttf',
"filter":
0,
"shuffle":
0,
"char_size":
80,
"canvas_size":
128,
"x_offset":
27,
"y_offset":
16,
"sample_count":
1000,
"sample_dir":
'C:/Users/1-16/python/neural-fonts-master/sample_dir',
"label":
0,
"fixed_sample":
0,
"all_sample":
0,
"handwriting_dir":
'C:/Users/1-16/python/neural-fonts-master/dst_dir'
})
if __name__ == "__main__":
hypothesis,
weights=(0.33, 0.33, 0.33, 0))
bleu_4 = corpus_bleu(references, hypothesis)
print('BLEU-1: {}, BLEU-2: {}, BLEU-3: {}, BLEU-4: {}'.format(
bleu_1, bleu_2, bleu_3, bleu_4))
print("val_loss is: {}".format(teller_val_losses.avg))
# Plot the BLEU Score and Validation Loss
if visualizer is not None:
visualizer.plot("BLEU_4", "val", iteration, bleu_4 * 100)
visualizer.plot("BLEU_3", "val", iteration, bleu_3 * 100)
visualizer.plot("BLEU_2", "val", iteration, bleu_2 * 100)
visualizer.plot("BLEU_1", "val", iteration, bleu_1 * 100)
print(type(teller_val_losses.avg.item()))
visualizer.plot("Teller Decoder Val Loss",
'val',
iteration,
teller_val_losses.avg.item(),
total=self.iterations)
self.model.predict(batch)
if __name__ == "__main__":
config_file = "example_args/gandraw_teller_args.json"
with open(config_file, 'r') as f:
cfg = json.load(f)
cfg = easydict.EasyDict(cfg)
tester = TellerTester(cfg, test_eval=True)
tester.test()
def main():
num_episodes = 300
replay_memory_init_size = 500
replay_memory_size = 50000
replay_step_limit = 100
batch_size = 128
discount_factor = 0.99
max_step = 200
replay_repeat = 5
update_target_per_steps = 100
tau = 0.0003
display = True
Transition = namedtuple("Transition",
"state action reward next_state done")
replay_memory = ReplayQueue(replay_memory_size)
env = gym.make('Pendulum-v0')
status_dim = env.observation_space.shape
action_dim = env.action_space.shape
if display:
env.reset()
env.render()
naf_cfg = easydict.EasyDict(
dict(status_dim=status_dim,
action_dim=action_dim,
hidden_dim=200,
learning_rate=0.0003,
l2_reg=0.0003,
is_training=True))
tf.reset_default_graph()
sess = tf.InteractiveSession()
date_str = datetime.now().strftime("%m%d_%H%M%S")
summaries_dir = os.path.abspath("./summary/naf/" + date_str)
q_estimator = NAF(naf_cfg, scope="q", summaries_dir=summaries_dir)
target_estimator = NAF(naf_cfg, sess=sess, scope="target")
sess.run(tf.global_variables_initializer())
copy_params(sess, q_estimator, target_estimator, 1)
saver = tf.train.Saver()
checkpoint_dir = os.path.abspath("./checkpoints/naf/" + date_str)
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
checkpoint_path = os.path.join(checkpoint_dir, "model")
# saver.restore(sess, "checkpoints/naf/XXXX_XXXXXX/model")
state = env.reset()
k = 0
for _ in range(replay_memory_init_size):
action = env.action_space.sample()
next_state, reward, done, _ = env.step(action)
replay_memory.push(Transition(state, action, reward, next_state, done))
if done or k >= replay_step_limit:
k = 0
state = env.reset()
else:
k += 1
state = next_state
for i_episode in range(1, num_episodes + 1):
state = env.reset()
episode_reward = 0
if display:
env.render()
for i_step in itertools.count(1):
action = q_estimator.predict(np.expand_dims(state, 0))[0]
action += np.random.randn(action_dim[0]) / i_episode # TODO
action = np.minimum(env.action_space.high,
np.maximum(env.action_space.low, action))
next_state, reward, done, _ = env.step(action)
if display:
env.render()
episode_reward += reward
replay_memory.push(
Transition(state, action, reward, next_state, done))
for _ in range(replay_repeat):
samples = replay_memory.sample(batch_size)
states_batch, action_batch, reward_batch, next_states_batch, done_batch =\
map(np.array, zip(*samples))
target_values = target_estimator.values(next_states_batch)
target_batch = reward_batch + (
1 - done_batch) * discount_factor * target_values
loss = q_estimator.update(states_batch, action_batch,
target_batch)
print(
"\rEpisode {}/{} step {} action {:.3f} reward {:.3f} loss {:.6f}"
.format(i_episode, num_episodes, i_step, action[0],
episode_reward, loss),
end='',
flush=True)
if i_step % update_target_per_steps == 0:
copy_params(sess, q_estimator, target_estimator,
tau * update_target_per_steps)
if done or i_step >= max_step:
print()
break
state = next_state
if i_episode % 50 == 0:
print('Saved!')
saver.save(sess, checkpoint_path)
if q_estimator.summary_writer:
episode_summary = tf.Summary()
episode_summary.value.add(simple_value=episode_reward,
node_name="episode_reward",
tag="episode_reward")
q_estimator.summary_writer.add_summary(episode_summary, i_episode)
q_estimator.summary_writer.flush()
saver.save(sess, checkpoint_path)
sess.close()
def __init__(self, opt):
super(LocalizationLayer, self).__init__()
self.opt = easydict.EasyDict()
self.opt.input_dim = utils.getopt(opt, 'input_dim')
self.opt.output_size = utils.getopt(opt, 'output_size')
# list x0, y0, sx, sy
self.opt.field_centers = utils.getopt(opt, 'field_centers')
self.opt.mid_box_reg_weight = utils.getopt(opt, 'mid_box_reg_weight')
self.opt.mid_objectness_weight = utils.getopt(opt,
'mid_objectness_weight')
self.opt.rpn_filter_size = utils.getopt(opt, 'rpn_filter_size', 3)
self.opt.rpn_num_filters = utils.getopt(opt, 'rpn_num_filters', 256)
self.opt.zero_box_conv = utils.getopt(opt, 'zero_box_conv', True)
self.opt.std = utils.getopt(opt, 'std', 0.01)
self.opt.anchor_scale = utils.getopt(opt, 'anchor_scale', 1.0)
self.opt.anchors = utils.getopt(opt, 'anchors', 'original')
self.opt.sampler_batch_size = utils.getopt(opt, 'sampler_batch_size',
256)
self.opt.sampler_high_thresh = utils.getopt(opt, 'sampler_high_thresh',
0.75)
self.opt.sampler_low_thresh = utils.getopt(opt, 'sampler_low_thresh',
0.4)
self.opt.train_remove_outbounds_boxes = utils.getopt(
opt, 'train_remove_outbounds_boxes', 1)
self.opt.box_reg_decay = utils.getopt(opt, 'box_reg_decay', 5e-5)
self.opt.tunable_anchors = utils.getopt(opt, 'tunable_anchors', False)
self.opt.backprop_rpn_anchors = utils.getopt(opt,
'backprop_rpn_anchors',
False)
self.box_loss = utils.getopt(opt, 'wordness_loss')
self.opt.contrastive_loss = utils.getopt(opt, 'contrastive_loss')
self.stats = easydict.EasyDict()
self.stats.losses = easydict.EasyDict()
self.stats.vars = easydict.EasyDict()
self.dtp_train = utils.getopt(opt, 'dtp_train', False)
if self.dtp_train:
self.opt.sampler_batch_size /= 2
sampler_opt = {
'batch_size': self.opt.sampler_batch_size,
'low_thresh': self.opt.sampler_low_thresh,
'high_thresh': self.opt.sampler_high_thresh,
'contrastive_loss': self.opt.contrastive_loss
}
debug_sampler = utils.getopt(opt, 'box_sampler', False)
if debug_sampler != False:
sampler_opt['box_sampler'] = debug_sampler
self.rpn = RPN(self.opt)
self.box_sampler_helper = BoxSamplerHelper(sampler_opt)
self.roi_pooling = BilinearRoiPooling(self.opt.output_size[0],
self.opt.output_size[1])
self.invert_box_transform = InvertBoxTransform()
# Construct criterions
if self.opt.backprop_rpn_anchors:
self.box_reg_loss = BoxRegressionCriterion(
self.opt.mid_box_reg_weight)
else:
self.box_reg_loss = nn.SmoothL1Loss() # for RPN box regression
self.box_scoring_loss = nn.CrossEntropyLoss()
self.image_height = None
self.image_width = None
self._called_forward_size = False
self._called_backward_size = False
def graph_forward(self, opt, batch, training=False):
var = easydict.EasyDict()
[var.image, var.points_GT] = [v.to(opt.device) for v in batch]
var.points_pred = self.network.forward(opt, var.image)
return var
def get_config(config):
with open(config, 'r') as stream:
return easydict.EasyDict(yaml.load(stream))
def main():
os.environ["MXNET_CUDNN_AUTOTUNE_DEFAULT"] = "0"
os.environ["MXNET_GPU_MEM_POOL_TYPE"] = "Round"
args = parse_args()
setattr(mobula.config, "NVCC", args.nvcc)
config = easydict.EasyDict()
config.gpus = [int(x) for x in str(args.gpus).split(',')]
config.dataset = easydict.EasyDict()
config.dataset.NUM_CLASSES = args.num_classes
config.dataset.dataset_type = args.dataset_type
config.dataset.dataset_path = args.dataset_root
config.retinanet = easydict.EasyDict()
config.retinanet.network = easydict.EasyDict()
config.retinanet.network.FPN_STRIDES = [8, 16, 32, 64, 128]
config.retinanet.network.BASE_SIZES = [(32, 32), (64, 64), (128, 128),
(256, 256), (512, 512)]
config.retinanet.network.SCALES = [2**0, 2**(1 / 2), 2**(2 / 3)]
config.retinanet.network.RATIOS = [1 / 2, 1, 2]
config.retinanet.network.bbox_norm_coef = [0.1, 0.1, 0.2, 0.2]
config.TRAIN = easydict.EasyDict()
config.TRAIN.batch_size = 2 * len(config.gpus)
config.TRAIN.lr = 0.01 * config.TRAIN.batch_size / 16
config.TRAIN.warmup_lr = config.TRAIN.lr
config.TRAIN.warmup_step = 1000
config.TRAIN.wd = 1e-4
config.TRAIN.momentum = .9
config.TRAIN.log_path = "output/{}/RetinaNet-hflip".format(
config.dataset.dataset_type, config.TRAIN.lr)
config.TRAIN.log_interval = 100
config.TRAIN.cls_focal_loss_alpha = .25
config.TRAIN.cls_focal_loss_gamma = 2
config.TRAIN.image_short_size = 600
config.TRAIN.image_max_long_size = 1333
config.TRAIN.aspect_grouping = True
# if aspect_grouping is set to False, all images will be pad to (PAD_H, PAD_W)
config.TRAIN.PAD_H = 768
config.TRAIN.PAD_W = 768
config.TRAIN.begin_epoch = 0
config.TRAIN.end_epoch = 28
config.TRAIN.lr_step = [6, 8]
config.TRAIN.FLIP = True
config.TRAIN.resume = None
config.TRAIN.trainer_resume = None
config.network = easydict.EasyDict()
config.network.FIXED_PARAMS = []
config.network.sync_bn = True
config.network.use_global_stats = False if config.network.sync_bn else True
config.network.merge_backbone_bn = False
config.val = easydict.EasyDict()
if args.demo:
config.val.params_file = args.demo_params
config.val.viz = args.viz
demo_net(config)
else:
os.makedirs(config.TRAIN.log_path, exist_ok=True)
log_init(filename=os.path.join(config.TRAIN.log_path,
"train_{}.log".format(time.time())))
msg = pprint.pformat(config)
logging.info(msg)
train_net(config)
""" """ import easydict config = easydict.EasyDict() config.tv_weight = 10 config.generator_g_lr = 1e-4 config.generator_f_lr = 1e-4 config.discriminator_c_lr = 1e-4 config.discriminator_t_lr = 1e-4 config.model_path = ''
def resultControl():
if(request.method == 'POST'):
# 클라이언트에서 sentenceId & wav file 받아옴
wav = request.files['receiveFile']
filename = request.form['fileName']
sentenceId = request.form['sentenceId']
# upload 디렉터리에 저장
wav.save(FILE_DIRECTORY + secure_filename(wav.filename))
##### upload 디렉터리에 있는 파일을 STT로 변한
# 임시 path
args = easydict.EasyDict({"local_file_path": "./uploadFile/"+filename})
# TODO Credential Error
# print(sample_recognize(args.local_file_path))
# sentenceId를 통해 DB에서 표준 발음 텍스트 가져옴
Pick_sentence = db_session.query(Sentence).filter(Sentence.sentenceId == sentenceId).first()
receiveSTTData = sample_recognize(args.local_file_path)
receiveData = similaritySentence(receiveSTTData, Pick_sentence.standard)
#receiveData = "날시가 참 말따"
print("STT result : ", receiveData)
# print(Pick_sentence)
##### 분석 알고리즘
hannanum = Hannanum()
StandardSentence = Pick_sentence.standard
sentenceData = Pick_sentence.sentenceData
# 공백 인덱스 리스트 생성
# 공백 개수는 일치
userBlankList = [i for i, value in enumerate(receiveData) if value == " "]
standardBlankList = [i for i, value in enumerate(StandardSentence) if value == " "]
# print(BlankList)
# 문자열 길이가 다르거나 공백 개수가 다르면
# 재시도 요청
if (len(receiveData) != len(StandardSentence) or len(userBlankList) != len(standardBlankList)):
os.remove("./uploadFile/"+filename)
return jsonify(
status="failure",
resultData=receiveData,
errorMessage="repeat",
)
# 공백 제거
UserSentence = receiveData.replace(" ", "")
StandardSentence = StandardSentence.replace(" ", "")
sentenceData = sentenceData.replace(" ", "")
# print(UserSentence)
# print(StandardSentence)
Total_pho = 0 # 총 음소 개수
Wrong_total_pho = 0 # 틀린 음소 개수
Wrong_word_index_list = [] # 틀린 글자 데이터가 들어있는 리스트
Wrong_word_list = [] # 틀린 단어 데이터가 들어있는 리스트
Wrong_pho_dict = {'u' : {},
'm' : {}, # 틀린 음소가 저장되는 딕셔너리
'b' : {}} # u : 자음, m : 모음, b : 받침
for index, standard in enumerate(StandardSentence):
StandPho = phonemeConvert(standard)
# 글자가 일치하는 경우
if(UserSentence[index] == standard):
if(StandPho[2] == ' '):
Total_pho += 2
else:
Total_pho += 3
# 글자가 일치하지 않는 경우
# 음소 분해
else:
Wrong_word_index_list.append(index)
UserPho = phonemeConvert(UserSentence[index])
SentencePho = phonemeConvert(sentenceData[index])
if(UserPho[2] == ' ' and StandPho[2] == ' '):
Total_pho += 2
else:
Total_pho += 3
if(UserPho[2] != StandPho[2]):
Wrong_total_pho += 1
if StandPho[2] != ' ':
if StandPho[2] in Wrong_pho_dict['b']:
Wrong_pho_dict['b'][SentencePho[2]] += 1
else:
Wrong_pho_dict['b'][SentencePho[2]] = 1
if (UserPho[0] != StandPho[0]):
Wrong_total_pho += 1
if StandPho[0] in Wrong_pho_dict['u']:
Wrong_pho_dict['u'][SentencePho[0]] += 1
else:
Wrong_pho_dict['u'][SentencePho[0]] = 1
if (UserPho[1] != StandPho[1]):
Wrong_total_pho += 1
if StandPho[1] in Wrong_pho_dict['m']:
Wrong_pho_dict['m'][SentencePho[1]] += 1
else:
Wrong_pho_dict['m'][SentencePho[1]] = 1
# print(Wrong_pho_dict)
######### 틀린 음소 record 테이블에 count 올림 -> TEST SUCCESS
for type in Wrong_pho_dict:
for pho in Wrong_pho_dict[type]:
# print(pho)
updateData = db_session.query(Record).filter(Record.recordType == type)\
.filter(Record.recordData == pho).first()
# print(updateData.type, updateData.recordData)
updateData.count += Wrong_pho_dict[type][pho]
db_session.commit()
# 일치율
Correct_rate = round(1 - (Wrong_total_pho / Total_pho), 4)
"""
# 일치율 100%인 경우
if Correct_rate == 1:
os.remove("./uploadFile/" + filename)
return jsonify(
status="perfect",
resultData=receiveData,
score=Correct_rate,
)
"""
# print(Wrong_word_list)
# 변경 후
# print(Wrong_word_index_list)
sentenceData_split = Pick_sentence.sentenceData.split()
# print(sentenceData_split)
# 틀린 인덱스가 포함된 단어 선택
word_start_point = 0
for sentence_word in sentenceData_split:
word_end_point = word_start_point + len(sentence_word)-1
# print(word_start_point, word_end_point)
for wrong_index in Wrong_word_index_list:
if word_start_point <= wrong_index and word_end_point >= wrong_index:
word_to_pos = hannanum.pos(sentence_word)
# print(word_to_pos)
wrong_word_pho_list = phonemeConvert(sentenceData[wrong_index])
# print(wrong_word_pho_list)
for pos in word_to_pos:
#TODO 틀린 단어에 N이나 P가 여러개 들어있으면??
if pos[1] == 'N' or pos[1] == 'P':
for pos_word in pos[0]:
pos_word_pho_list = phonemeConvert(pos_word)
# print(pos_word_pho_list)
if wrong_word_pho_list[0] == pos_word_pho_list[0]:
Wrong_word_list.append(pos)
break
word_start_point += len(sentence_word)
print(Wrong_word_list)
# 틀린 글자 인덱스를 원래 문장을 기준으로 변경
for i in userBlankList:
for index, j in enumerate(Wrong_word_index_list):
if(j >= i):
Wrong_word_index_list[index] += 1
# print(Wrong_word_index)
######## result 테이블에 결과값 저장 -> TEST SUCCESS
resultData = Result(stid=sentenceId, rsdata=receiveData, score=Correct_rate)
db_session.add(resultData)
db_session.commit()
# 일치율 100%인 경우
if Correct_rate == 1:
os.remove("./uploadFile/" + filename)
return jsonify(
status="perfect",
resultData=receiveData,
score=Correct_rate,
)
######## 가장 많이 틀린 단어에 대한 추천 문장 1개
recommend_OtoD = dict(sentenceId=-1, sentenceData="", standard="")
recommend_word = ""
# 틀린 단어 리스트에 단어가 존재할 경우
if Wrong_word_list:
random.shuffle(Wrong_word_list)
for random_select_word in Wrong_word_list:
Word_query = db_session.query(Word).filter(Word.wordData == random_select_word[0])\
.filter(Word.wordType == random_select_word[1]).filter(Word.sentenceId != sentenceId)
Word_entry = [pq.sentenceId for pq in Word_query]
if Word_entry:
recommend_word = random_select_word[0]
if random_select_word[1] == 'P':
recommend_word += '다'
random_select_setencdId = random.choice(Word_entry)
Recommend_sentence = db_session.query(Sentence).filter(Sentence.sentenceId == random_select_setencdId).first()
recommend_OtoD['sentenceId'] = Recommend_sentence.sentenceId
recommend_OtoD['sentenceData'] = Recommend_sentence.sentenceData
recommend_OtoD['standard'] = Recommend_sentence.standard
break
os.remove("./uploadFile/" + filename)
# response해줄 틀린 단어 리스트
wordList = []
for w in Wrong_word_list:
if w[1] == "P":
wordList.append(w[0]+"다")
else:
wordList.append(w[0])
# 결과 데이터를 모두 json으로 묶음
return jsonify(
status = "success",
score = Correct_rate,
wordList = wordList,
userBlank = userBlankList,
standardBlank = standardBlankList,
wrongIndex = Wrong_word_index_list,
resultData = receiveData
)
# -*- coding:utf-8 -*- # author:平手友梨奈ii # e-mail:[email protected] # datetime:1993/12/01 # filename:configs.py # software: PyCharm import easydict CONFIG = easydict.EasyDict() # mAP evaluation CONFIG.DETECT = easydict.EasyDict() CONFIG.DETECT.SCORE = 0.3 CONFIG.DETECT.IOU = 0.5 CONFIG.DETECT.RESOLUTION = (416, 416) CONFIG.DETECT.mAP_THRES = 0.5 # prediction CONFIG.PREDICT = easydict.EasyDict() CONFIG.PREDICT.WEIGHTS = 'logs/yolo4_tiny_weight.h5' CONFIG.PREDICT.ANCHOR_PATH = 'model_data/yolo4_tiny_anchors.txt' CONFIG.PREDICT.CLASS_PATH = 'model_data/coco_classes.txt' CONFIG.PREDICT.SCORE = 0.2 CONFIG.PREDICT.IOU = 0.3 CONFIG.PREDICT.RESOLUTION = (416, 416) CONFIG.PREDICT.MAX_BOXES = 40 # train CONFIG.TRAIN = easydict.EasyDict()
import multiResData
import PoseTools as pt
import easydict
res = pt.pickle_load(res_file)[0]
ims,locs,info = multiResData.read_and_decode_without_session(db,5,())
ims = np.array(ims)
locs = np.array(locs)
pred = res[-1][0]
dd = np.sqrt(np.sum((res[-1][0]-res[-1][1])**2,axis=-1))
gg = np.percentile(dd,[50,75,90,96],axis=0)
pt.show_result_hist(ims[0,...],locs[0,...],gg)
ndx = np.random.choice(ims.shape[0],6)
pt.show_result(ims,ndx,locs,pred)
conf = easydict.EasyDict()
conf.adjust_contrast = True
conf.clahe_grid_size = 20
cims = pt.adjust_contrast(ims,conf)
pt.show_result(cims,ndx,locs,pred)
## Original DLC and Leap Training
import run_apt_expts as rae
dtypes = ['alice','stephen']
for dd in dtypes:
reload(rae)
rae.setup(dd)
rae.train_deepcut_orig()
rae.train_leap_orig()
# In[ ]:
import argparse
from PIL import Image #This may not be used
import os ##This may not be used
import easydict
args = easydict.EasyDict({
'type': 'gan',
'lr': 0.0002,
'l1_coef': 50,
'l2_coef': 100,
'cls': True,
'save_path': './flowers_cls_test',
'inference': True,
'pre_trained_disc': 'checkpoints/flowers_cls/disc_190.pth',
'pre_trained_gen': 'checkpoints/flowers_cls/gen_190.pth',
'dataset': 'flowers',
'split': 2,
'batch_size': 64,
'num_workers': 8,
'epochs': 200
})
trainer = Trainer(type=args.type,
dataset=args.dataset,
split=args.split,
lr=args.lr,
save_path=args.save_path,
l1_coef=args.l1_coef,
l2_coef=args.l2_coef,
"wifi_funny_special": dummy_handler,
"network_percent_data": dummy_handler,
"battery_mean_usage": dummy_handler
}
user_achievement_handlers = {
"count_based": count_based_badge,
"procent_based": proc_based_badge,
"time_based": time_based_badge,
"wifi_security_special": wifi_security_special_badge,
"wifi_funny_special": wifi_funny_special_badge,
"network_percent_data": network_percent_data_badge,
"battery_mean_usage": battery_mean_usage_badge
}
user_ranking_handlers = {
"count_based": count_based_place,
"procent_based": proc_based_place,
"time_based": time_based_place,
"wifi_security_special": dummy_handler,
"wifi_funny_special": dummy_handler,
"network_percent_data": dummy_handler,
"battery_mean_usage": dummy_handler
}
handlers = easydict.EasyDict({})
handlers.ranking = ranking_handlers
handlers.user = {}
handlers.user.achievements = user_achievement_handlers
handlers.user.ranking = user_ranking_handlers