# -*- coding:utf-8 -*- # Created Time: 2017/11/04 21:19:32 # Author: Taihong Xiao <xiaotaihong@126.com> import tensorflow as tf import numpy as np import os from scipy import misc from model import Model from dataset import Dataset import argparse def get_representation(img, model, model_dir, out_path): saver = tf.train.Saver() with tf.Session() as sess: sess.run(tf.global_variables_initializer()) ckpt = tf.train.get_checkpoint_state(model_dir) # print(ckpt) # print(ckpt.model_checkpoint_path) if ckpt and ckpt.model_checkpoint_path: saver.restore(sess, ckpt.model_checkpoint_path) Ax = tf.placeholder(tf.float32, [model.batch_size,model.height,model.width,model.channel],name='Ax') enc_Ax = model.splitter('encoder', Ax) grad_att_1 =[tf.gradients(enc_Ax[0][:,:,:,i], Ax)[0] for i in range(128)] grad_att_2 =[tf.gradients(enc_Ax[1][:,:,:,i], Ax)[0] for i in range(128)] grad_att_3 =[tf.gradients(enc_Ax[2][:,:,:,i], Ax)[0] for i in range(256)] # from IPython import embed;embed();exit() grad_1 = sess.run(grad_att_1, feed_dict={Ax: img}) grad_2 = sess.run(grad_att_2, feed_dict={Ax: img}) grad_3 = sess.run(grad_att_3, feed_dict={Ax: img}) for i in range(128): misc.imsave(os.path.join(out_path, '0_{:03d}.jpg'.format(i)), grad_1[i][0]) misc.imsave(os.path.join(out_path, '1_{:03d}.jpg'.format(i)), grad_2[i][0]) np.save(os.path.join(out_path, '0_{:03d}.npy'.format(i)), grad_2[i][0]) np.save(os.path.join(out_path, '1_{:03d}.npy'.format(i)), grad_2[i][0]) for i in range(256): misc.imsave(os.path.join(out_path, '2_{:03d}.jpg'.format(i)), grad_3[i][0]) np.save(os.path.join(out_path, '2_{:03d}.npy'.format(i)), grad_3[i][0]) def main(): parser = argparse.ArgumentParser(description='test', formatter_class=argparse.RawTextHelpFormatter) parser.add_argument('-i', '--input', type=str, help='image path') parser.add_argument('-a', '--attributes', type=str, nargs='+', help='attribute list') parser.add_argument('--model_dir', type=str, default='train_log/model/', help='path to the model') parser.add_argument('--latent_path', type=str, default='latent', help='path to the model') parser.add_argument('-g', '--gpu', type=str, default='', help='gpu ids') args = parser.parse_args() os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu if not os.path.exists(args.latent_path): os.makedirs(args.latent_path) DNA_GAN = Model(feature_list=args.attributes, is_train=False, nhwc=[1,64,64,3]) img = np.expand_dims(misc.imresize(misc.imread(args.input), (DNA_GAN.height, DNA_GAN.width)), axis=0) get_representation(img, DNA_GAN, args.model_dir, args.latent_path) if __name__ == "__main__": main()