tensorflow65 《深度学习原理与TensorFlow实战》04 CNN看懂世界 02 vgg_network.py

基本环境

#《深度学习原理与TensorFlow实战》04 CNN看懂世界 02 vgg_network.py
# 书源码地址:https://github.com/DeepVisionTeam/TensorFlowBook.git
# 视频讲座地址：http://edu.csdn.net/course/detail/5222
# win10 Tensorflow1.2.0 python3.6.1
# CUDA v8.0 cudnn-8.0-windows10-x64-v5.1
# https://github.com/tflearn/tflearn/blob/master/examples/images/vgg_network.py
# vgg_network占用内存较大，看python内存占用大约8GB+，本机gpu为4GB，不够用，所以采用cpu only版本做测试。速度较慢。
# flowers17的数据会自动下载，且速度较快。
# 硬件配置：
# Intel(R) Xeon(R) CPU E3-1505M v5 @ 2.80GHz 2.81GHz
# 内存 64GB
# 系统： win10 x64
# 大约24小时内跑50+个epoch

vgg_network.py

# -*- coding: utf-8 -*-

""" Very Deep Convolutional Networks for Large-Scale Visual Recognition.

Applying VGG 16-layers convolutional network to Oxford's 17 Category Flower
Dataset classification task.

References:
    Very Deep Convolutional Networks for Large-Scale Image Recognition.
    K. Simonyan, A. Zisserman. arXiv technical report, 2014.

Links:
    http://arxiv.org/pdf/1409.1556

"""

from __future__ import division, print_function, absolute_import

import tflearn
from tflearn.layers.core import input_data, dropout, fully_connected
from tflearn.layers.conv import conv_2d, max_pool_2d
from tflearn.layers.estimator import regression

# Data loading and preprocessing
import tflearn.datasets.oxflower17 as oxflower17
X, Y = oxflower17.load_data(one_hot=True, resize_pics=(224, 224))

# Building 'VGG Network'
network = input_data(shape=[None, 224, 224, 3])

network = conv_2d(network, 64, 3, activation='relu')
network = conv_2d(network, 64, 3, activation='relu')
network = max_pool_2d(network, 2, strides=2)

network = conv_2d(network, 128, 3, activation='relu')
network = conv_2d(network, 128, 3, activation='relu')
network = max_pool_2d(network, 2, strides=2)

network = conv_2d(network, 256, 3, activation='relu')
network = conv_2d(network, 256, 3, activation='relu')
network = conv_2d(network, 256, 3, activation='relu')
network = max_pool_2d(network, 2, strides=2)

network = conv_2d(network, 512, 3, activation='relu')
network = conv_2d(network, 512, 3, activation='relu')
network = conv_2d(network, 512, 3, activation='relu')
network = max_pool_2d(network, 2, strides=2)

network = conv_2d(network, 512, 3, activation='relu')
network = conv_2d(network, 512, 3, activation='relu')
network = conv_2d(network, 512, 3, activation='relu')
network = max_pool_2d(network, 2, strides=2)

network = fully_connected(network, 4096, activation='relu')
network = dropout(network, 0.5)
network = fully_connected(network, 4096, activation='relu')
network = dropout(network, 0.5)
network = fully_connected(network, 17, activation='softmax')

network = regression(network, optimizer='rmsprop',
                     loss='categorical_crossentropy',
                     learning_rate=0.0001)

# Training
model = tflearn.DNN(network, checkpoint_path='model_vgg',
                    max_checkpoints=1, tensorboard_verbose=0)
model.fit(X, Y, n_epoch=500, shuffle=True,
          show_metric=True, batch_size=32, snapshot_step=500,
          snapshot_epoch=False, run_id='vgg_oxflowers17')
'''
Run id: vgg_oxflowers17
Log directory: /tmp/tflearn_logs/
---------------------------------
Training samples: 1360
Validation samples: 0
--
Training Step: 1  | time: 39.151s
| RMSProp | epoch: 001 | loss: 0.00000 - acc: 0.0000 -- iter: 0032/1360
Training Step: 2  | total loss: 2.55066 | time: 78.160s
| RMSProp | epoch: 001 | loss: 2.55066 - acc: 0.0562 -- iter: 0064/1360
Training Step: 3  | total loss: 2.78507 | time: 117.662s
| RMSProp | epoch: 001 | loss: 2.78507 - acc: 0.0358 -- iter: 0096/1360
Training Step: 4  | total loss: 2.82211 | time: 163.628s
| RMSProp | epoch: 001 | loss: 2.82211 - acc: 0.0558 -- iter: 0128/1360
Training Step: 5  | total loss: 2.83343 | time: 212.448s
| RMSProp | epoch: 001 | loss: 2.83343 - acc: 0.0172 -- iter: 0160/1360
Training Step: 6  | total loss: 2.82952 | time: 257.656s
| RMSProp | epoch: 001 | loss: 2.82952 - acc: 0.0664 -- iter: 0192/1360
Training Step: 7  | total loss: 2.83349 | time: 306.941s
| RMSProp | epoch: 001 | loss: 2.83349 - acc: 0.0266 -- iter: 0224/1360
Training Step: 8  | total loss: 2.83455 | time: 355.456s
| RMSProp | epoch: 001 | loss: 2.83455 - acc: 0.0468 -- iter: 0256/1360
Training Step: 9  | total loss: 2.83402 | time: 402.628s
| RMSProp | epoch: 001 | loss: 2.83402 - acc: 0.0716 -- iter: 0288/1360
Training Step: 10  | total loss: 2.83300 | time: 449.701s
| RMSProp | epoch: 001 | loss: 2.83300 - acc: 0.0514 -- iter: 0320/1360
Training Step: 11  | total loss: 2.83271 | time: 495.893s
| RMSProp | epoch: 001 | loss: 2.83271 - acc: 0.0863 -- iter: 0352/1360
Training Step: 12  | total loss: 2.83074 | time: 537.937s
| RMSProp | epoch: 001 | loss: 2.83074 - acc: 0.0615 -- iter: 0384/1360
Training Step: 13  | total loss: 2.83104 | time: 580.727s
| RMSProp | epoch: 001 | loss: 2.83104 - acc: 0.0619 -- iter: 0416/1360
Training Step: 14  | total loss: 2.83621 | time: 623.209s
| RMSProp | epoch: 001 | loss: 2.83621 - acc: 0.0366 -- iter: 0448/1360
Training Step: 15  | total loss: 2.83518 | time: 664.900s
| RMSProp | epoch: 001 | loss: 2.83518 - acc: 0.0345 -- iter: 0480/1360
Training Step: 16  | total loss: 2.83590 | time: 706.464s
| RMSProp | epoch: 001 | loss: 2.83590 - acc: 0.0333 -- iter: 0512/1360
Training Step: 17  | total loss: 2.83771 | time: 746.936s
| RMSProp | epoch: 001 | loss: 2.83771 - acc: 0.0213 -- iter: 0544/1360
Training Step: 18  | total loss: 2.83808 | time: 788.010s
| RMSProp | epoch: 001 | loss: 2.83808 - acc: 0.0139 -- iter: 0576/1360
Training Step: 19  | total loss: 2.83710 | time: 829.101s
| RMSProp | epoch: 001 | loss: 2.83710 - acc: 0.0093 -- iter: 0608/1360
Training Step: 20  | total loss: 2.83573 | time: 869.866s
| RMSProp | epoch: 001 | loss: 2.83573 - acc: 0.0163 -- iter: 0640/1360
Training Step: 21  | total loss: 2.83293 | time: 910.728s
| RMSProp | epoch: 001 | loss: 2.83293 - acc: 0.0307 -- iter: 0672/1360
Training Step: 22  | total loss: 2.83208 | time: 951.732s
| RMSProp | epoch: 001 | loss: 2.83208 - acc: 0.0496 -- iter: 0704/1360
Training Step: 23  | total loss: 2.83391 | time: 992.608s
| RMSProp | epoch: 001 | loss: 2.83391 - acc: 0.0443 -- iter: 0736/1360
Training Step: 24  | total loss: 2.83574 | time: 1033.661s
| RMSProp | epoch: 001 | loss: 2.83574 - acc: 0.0406 -- iter: 0768/1360
Training Step: 25  | total loss: 2.83504 | time: 1074.329s
| RMSProp | epoch: 001 | loss: 2.83504 - acc: 0.0466 -- iter: 0800/1360
Training Step: 26  | total loss: 2.83532 | time: 1115.530s
| RMSProp | epoch: 001 | loss: 2.83532 - acc: 0.0342 -- iter: 0832/1360
Training Step: 27  | total loss: 2.83454 | time: 1156.374s
| RMSProp | epoch: 001 | loss: 2.83454 - acc: 0.0335 -- iter: 0864/1360
Training Step: 28  | total loss: 2.83370 | time: 1196.976s
| RMSProp | epoch: 001 | loss: 2.83370 - acc: 0.0329 -- iter: 0896/1360
Training Step: 29  | total loss: 2.83326 | time: 1237.924s
| RMSProp | epoch: 001 | loss: 2.83326 - acc: 0.0401 -- iter: 0928/1360
Training Step: 30  | total loss: 2.83426 | time: 1278.819s
| RMSProp | epoch: 001 | loss: 2.83426 - acc: 0.0380 -- iter: 0960/1360
Training Step: 31  | total loss: 2.83360 | time: 1319.769s
| RMSProp | epoch: 001 | loss: 2.83360 - acc: 0.0365 -- iter: 0992/1360
Training Step: 32  | total loss: 2.83191 | time: 1360.613s
| RMSProp | epoch: 001 | loss: 2.83191 - acc: 0.0493 -- iter: 1024/1360
Training Step: 33  | total loss: 2.83120 | time: 1401.292s
| RMSProp | epoch: 001 | loss: 2.83120 - acc: 0.0591 -- iter: 1056/1360
Training Step: 34  | total loss: 2.83001 | time: 1442.211s
| RMSProp | epoch: 001 | loss: 2.83001 - acc: 0.0665 -- iter: 1088/1360
Training Step: 35  | total loss: 2.82995 | time: 1482.782s
| RMSProp | epoch: 001 | loss: 2.82995 - acc: 0.0591 -- iter: 1120/1360
Training Step: 36  | total loss: 2.83233 | time: 1523.651s
| RMSProp | epoch: 001 | loss: 2.83233 - acc: 0.0534 -- iter: 1152/1360
Training Step: 37  | total loss: 2.83467 | time: 1564.673s
| RMSProp | epoch: 001 | loss: 2.83467 - acc: 0.0552 -- iter: 1184/1360
Training Step: 38  | total loss: 2.83420 | time: 1605.479s
| RMSProp | epoch: 001 | loss: 2.83420 - acc: 0.0506 -- iter: 1216/1360
Training Step: 39  | total loss: 2.83358 | time: 1646.481s
| RMSProp | epoch: 001 | loss: 2.83358 - acc: 0.0528 -- iter: 1248/1360
Training Step: 40  | total loss: 2.83449 | time: 1687.051s
| RMSProp | epoch: 001 | loss: 2.83449 - acc: 0.0547 -- iter: 1280/1360
Training Step: 41  | total loss: 2.83445 | time: 1727.702s
| RMSProp | epoch: 001 | loss: 2.83445 - acc: 0.0504 -- iter: 1312/1360
Training Step: 42  | total loss: 2.83419 | time: 1768.542s
| RMSProp | epoch: 001 | loss: 2.83419 - acc: 0.0525 -- iter: 1344/1360
Training Step: 43  | total loss: 2.83316 | time: 1789.223s
| RMSProp | epoch: 001 | loss: 2.83316 - acc: 0.0543 -- iter: 1360/1360
Training Step: 44  | total loss: 2.83455 | time: 20.706s
| RMSProp | epoch: 002 | loss: 2.83455 - acc: 0.0449 -- iter: 0032/1360
Training Step: 45  | total loss: 2.83637 | time: 61.555s
| RMSProp | epoch: 002 | loss: 2.83637 - acc: 0.0373 -- iter: 0064/1360
Training Step: 46  | total loss: 2.83504 | time: 102.520s
| RMSProp | epoch: 002 | loss: 2.83504 - acc: 0.0467 -- iter: 0096/1360
Training Step: 47  | total loss: 2.83507 | time: 143.075s
| RMSProp | epoch: 002 | loss: 2.83507 - acc: 0.0390 -- iter: 0128/1360
Training Step: 48  | total loss: 2.83601 | time: 184.085s
...
| RMSProp | epoch: 059 | loss: 0.07563 - acc: 0.9953 -- iter: 0448/1360
Training Step: 2509  | total loss: 0.06807 | time: 545.071s
| RMSProp | epoch: 059 | loss: 0.06807 - acc: 0.9957 -- iter: 0480/1360
Training Step: 2510  | total loss: 0.06129 | time: 584.407s
| RMSProp | epoch: 059 | loss: 0.06129 - acc: 0.9962 -- iter: 0512/1360
Training Step: 2511  | total loss: 0.05854 | time: 623.157s
| RMSProp | epoch: 059 | loss: 0.05854 - acc: 0.9934 -- iter: 0544/1360
Training Step: 2512  | total loss: 0.05310 | time: 663.058s
| RMSProp | epoch: 059 | loss: 0.05310 - acc: 0.9941 -- iter: 0576/1360
Training Step: 2513  | total loss: 0.05105 | time: 702.142s
| RMSProp | epoch: 059 | loss: 0.05105 - acc: 0.9947 -- iter: 0608/1360
'''