基于2D卷积与3D卷积的多张侧脸重建正脸算法python源码+运行说明.zip

# -*- coding: utf-8 -*- from __future__ import division from __future__ import print_function import logging import pickle import sys from time import localtime, strftime from net_input_everything_featparts4 import * from ops import * from utils import * relu = tf.nn.relu import tensorflow as tf import os import heapq from sklearn.decomposition import PCA import Test2 as T import cell as C import os.path import time import numpy as np import shutil import face_recognition import os # X = np.array([[-1,2,66,-1], [-2,6,58,-1], [-3,8,45,-2], [1,9,36,1], [2,10,62,1], [3,5,83,2]]) #导入数据，维度为4 # pca = PCA(n_components=2) #降到2维 # pca.fit(X) #训练 # newX=pca.fit_transform(X) #降维后的数据 # 设置 GPU 按需增长 config = tf.ConfigProto() config.gpu_options.allow_growth = True # sess = tf.Session(config=config) os.system('echo $PATH') os.environ["CUDA_VISIBLE_DEVICES"] = "0,1" #指定前二块GPU可用 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' # doesn’t enable AVX/FMA #These parameters should provide a good initialization, but if for specific refinement, you can adjust them during training. # SYM_W = 0.3 #lamana1 ALPHA_ADVER = 1 #lamana2 BELTA_FEATURE = 1e-3 ##lamana3 # FEATURE = 1e-3 TV_WEIGHT = 1e-3 #lamana4 # IDEN_W = 1e-3 #a er fa UPDATE_G = 1 #optimize D once and UPDATE_G times G UPDATE_D = 1 MODE = 'fs60' #'f' feature loss enabled. 'v' -verification enanbled. 'o' original, 'm' masked is mandatory and no need to specify LOAD_60_LABEL = False #otherwise load frontal label WITHOUT_CODEMAP = True USE_MASK = False RANDOM_VERIFY = False CHANNEL = 3 ########################################################### ############################################################ flags = tf.app.flags flags.DEFINE_integer("epoch", 300, "Epoch to train [250]") flags.DEFINE_float("learning_rate", 2e-4, "Learning rate of for adam [0.0002]") flags.DEFINE_float("beta1", 0.5, "Momentum term of adam [0.5]") #flags.DEFINE_integer("train_size", np.inf, "The size of train images [np.inf]") flags.DEFINE_float("train_size", np.inf, "The size of train images [np.inf]") flags.DEFINE_integer("batch_size", 1, "The size of batch images [64]") flags.DEFINE_integer("image_size", 128, "The size of image to use (will be center cropped) [108]") flags.DEFINE_integer("output_size", 128, "The size of the output images to produce [64]") flags.DEFINE_integer("c_dim", 3, "Dimension of image color. [3]") flags.DEFINE_string("dataset", "MultiPIE", "The name of dataset [celebA, mnist, lsun]") flags.DEFINE_string("checkpoint_dir", "data1/check_3_17", "Directory name to save the checkpoints [checkpoint]") flags.DEFINE_string("sample_dir", "data1/samples", "Directory name to save the image samples [samples]") flags.DEFINE_boolean("is_train", True, "True for training, False for testing [False]") flags.DEFINE_boolean("is_crop", True, "True for training, False for testing [False]") flags.DEFINE_boolean("visualize", True, "True for visualizing, False for nothing [False]") FLAGS = flags.FLAGS #tf.subtract(x, y, name=None) # 减法 class DCGAN(object): def __init__(self, sess,batch_size=10,output_size=128, gf_dim=48, df_dim=48, dataset_name='MultiPIE',checkpoint_dir=None): """ Args: sess: TensorFlow session batch_size: The size of batch. Should be specified before training. output_size: (optional) The resolution in pixels of the images. [64] y_dim: (optional) Dimension of dim for y. [None] z_dim: (optional) Dimension of dim for Z. [100] gf_dim: (optional) Dimension of gen filters in first conv layer. [64] df_dim: (optional) Dimension of discrim filters in first conv layer. [64] gfc_dim: (optional) Dimension of gen units for for fully connected layer. [1024] dfc_dim: (optional) Dimension of discrim units for fully connected layer. [1024] c_dim: (optional) Dimension of image color. For grayscale input, set to 1. [3] """ self.flage = False self.testing = False # False True self.sess = sess self.batch_size = 2 self.time_size = 6 self.lstm = True self.Positions = {15: 0, 30: 1, 45: 2, 60: 3, 75: 4, 90: 5} self.test_batch_size = self.batch_size ## 5 self.output_size = output_size ## 128 self.gf_dim = gf_dim #眼睛使用的内核深度:64 ##生成器中第一个卷积层的filter维度 self.df_dim = df_dim #???64 ##鉴别器中第一个卷积层的filter维度 self.z_dim = 100 random.seed() self.DeepFacePath = '/media/gpu/文档/lsf/TP-GAN/DeepFace168.pickle' self.dataset_name = dataset_name #MultiPIE self.checkpoint_dir = checkpoint_dir #None self.loadDeepFace(self.DeepFacePath) # self.G = np.empty([2,128,128,3]) # self.labels = np.empty([2, 128, 128, 3]) def build_model(self): self.images_with_code2 = tf.placeholder(tf.float32, [self.batch_size] + [self.time_size] + [self.output_size, self.output_size, CHANNEL], name='images_with_code') self.labels = tf.placeholder(tf.float32, [self.batch_size] + [self.output_size, self.output_size, CHANNEL], name='label_images') self.g32_labels = tf.image.resize_bilinear(self.labels, [32, 32]) #双线性插值法 float32 (10,32,32,3) self.g64_labels = tf.image.resize_bilinear(self.labels, [64, 64]) #双线性插值法 float32 (10,64,64,3) # self.z = tf.random_normal([self.batch_size, 128,128,2], mean=0.0, stddev=0.02, seed=2017) # (10,100)正太分布 self.z = tf.random_normal([self.batch_size, 128, 128, 1], mean=0.0, stddev=0.02, seed=2017) # (10,100)正太分布 self.feats_lstms = [] self.feats2D = [] self.feats3D = [] self.img128 = [] self.g_loss = self.d_loss = 0 for i in range(self.time_size): self.featsfus = [] if i == 0: reuse = False else: reuse = True self.feats = self.generator(self.images_with_code2[:,i,...], name="encoder_2D",reuse=reuse) self.feats = list(self.feats) self.feats_lstm = [] for j in range(len(self.feats)): feats_0 = tf.reshape(self.feats[j],[self.feats[j].shape[0], 1, self.feats[j].shape[1], self.feats[j].shape[2],self.feats[j].shape[3]]) if i == 0: self.feats_lstm.append(feats_0) if i > 0: feats_1 = tf.concat([feats_0, self.feats_lstms[i - 1][j]], 1) self.feats_lstm.append(feats_1) self.feats_lstms.append(self.feats_lstm) if i == 0: self.fts = self.feats continue if i == 1: reuse = False # a = tf.image.rgb_to_grayscale(self.images_with_code2[:, i - 1, ...]) # b = tf.image.rgb_to_grayscale(self.images_with_code2[:, i, ...]) # c = tf.reduce_mean(tf.concat([self.images_with_code2[:, i - 1, ...], self.images_with_code2[:, i, ...]], -1), -1,keepdims=True) # self.G = tf.concat([a, c, b], -1) self.G = self.images_with_code2[:, i-1, ...] else: reuse = True h_state_0 = self.network2(self.feats_lstms[-1][0], "network2_h0",reuse=reuse) h_state_1 = self.network2(self.feats_lstms[-1][1], "network2_h1",reuse=reuse) h_state_2 = self.network2(self.feats_lstms[-1][2], "network2_h2",reuse=reuse) h_state_3 = self.network2(self.feats_lstms[-1][3], "network2_h3",reuse=reuse) # (6,16,16,256) self.feats_2D = h_state_0, h_state_1, h_state_2, h_state_3# self.h_state_5 self.feats2D.append(self.feats_2D) self.fts_feats = [] for m in ran