딥러닝 텐서플로우 개와 고양이 분류하는 Neural Network 만들기

Using more sophisticated images with Convolutional Neural Networks¶

I실생활의 이미지는 모양도 다르고, 비율도 다르고, 색깔도 다양하다. 이러한 것들을 분류하는 CNN을 만들어 본다.

1. Cats and Dogs 이미지 이용
2. 개와 고양이 분류하는 Neural Network 만들기
3. Evaluate the Training and Validation accuracy

Explore the Example Data¶

/tmp 컬럼에, 2000개의 이미지를 다운로드 받아서 저장한다.

NOTE: 2,000 개의 이미지는 캐글에서 가져왔다. "Dogs vs. Cats" dataset 원래는 25,000 개의 이미지 이지만, 실습용으로 추렸음.

In [ ]:

In [2]:

!wget --no-check-certificate \
  https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip \
  -O /tmp/cats_and_dogs_filtered.zip

--2022-12-30 05:22:44--  https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip
Resolving storage.googleapis.com (storage.googleapis.com)... 172.217.194.128, 74.125.200.128, 74.125.68.128, ...
Connecting to storage.googleapis.com (storage.googleapis.com)|172.217.194.128|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 68606236 (65M) [application/zip]
Saving to: ‘/tmp/cats_and_dogs_filtered.zip’

/tmp/cats_and_dogs_ 100%[===================>]  65.43M  21.3MB/s    in 3.9s    

2022-12-30 05:22:48 (16.9 MB/s) - ‘/tmp/cats_and_dogs_filtered.zip’ saved [68606236/68606236]

압축 풀기¶

In [3]:

import zipfile

In [4]:

file = zipfile.ZipFile('/tmp/cats_and_dogs_filtered.zip')

In [5]:

file.extractall('/tmp')

In [ ]:

데이터 억세스할 경로를 만든다.¶

In [6]:

base_dir = '/tmp/cats_and_dogs_filtered'

In [7]:

train_dir = '/tmp/cats_and_dogs_filtered/train'

In [8]:

test_dir =  '/tmp/cats_and_dogs_filtered/validation'

In [ ]:

파일명을 찍어본다.¶

In [ ]:

# 학습용 개 사진이 저장된 디렉토리 안에 있는 파일명을 확인

In [9]:

import os

In [11]:

os.listdir(train_dir+'/dogs')

Out[11]:

['dog.370.jpg',
 'dog.387.jpg',
 'dog.568.jpg',
 'dog.694.jpg',
 'dog.395.jpg',
 'dog.436.jpg',
 'dog.106.jpg',
 'dog.560.jpg',
 'dog.580.jpg',
 'dog.438.jpg',
 'dog.801.jpg',
 'dog.469.jpg',
 'dog.961.jpg',
 'dog.525.jpg',
 'dog.409.jpg',
 'dog.292.jpg',
 'dog.715.jpg',
 'dog.495.jpg',
 'dog.453.jpg',
 'dog.80.jpg',
 'dog.698.jpg',
 'dog.796.jpg',
 'dog.586.jpg',
 'dog.52.jpg',
 'dog.724.jpg',
 'dog.269.jpg',
 'dog.602.jpg',
 'dog.550.jpg',
 'dog.434.jpg',
 'dog.723.jpg',
 'dog.147.jpg',
 'dog.413.jpg',
 'dog.827.jpg',
 'dog.522.jpg',
 'dog.42.jpg',
 'dog.420.jpg',
 'dog.247.jpg',
 'dog.833.jpg',
 'dog.25.jpg',
 'dog.270.jpg',
 'dog.883.jpg',
 'dog.280.jpg',
 'dog.819.jpg',
 'dog.745.jpg',
 'dog.786.jpg',
 'dog.686.jpg',
 'dog.893.jpg',
 'dog.863.jpg',
 'dog.81.jpg',
 'dog.151.jpg',
 'dog.784.jpg',
 'dog.824.jpg',
 'dog.552.jpg',
 'dog.18.jpg',
 'dog.907.jpg',
 'dog.481.jpg',
 'dog.273.jpg',
 'dog.514.jpg',
 'dog.748.jpg',
 'dog.492.jpg',
 'dog.205.jpg',
 'dog.296.jpg',
 'dog.993.jpg',
 'dog.13.jpg',
 'dog.675.jpg',
 'dog.235.jpg',
 'dog.298.jpg',
 'dog.230.jpg',
 'dog.204.jpg',
 'dog.895.jpg',
 'dog.974.jpg',
 'dog.344.jpg',
 'dog.868.jpg',
 'dog.349.jpg',
 'dog.288.jpg',
 'dog.428.jpg',
 'dog.427.jpg',
 'dog.847.jpg',
 'dog.322.jpg',
 'dog.339.jpg',
 'dog.468.jpg',
 'dog.19.jpg',
 'dog.902.jpg',
 'dog.783.jpg',
 'dog.197.jpg',
 'dog.76.jpg',
 'dog.435.jpg',
 'dog.478.jpg',
 'dog.251.jpg',
 'dog.926.jpg',
 'dog.43.jpg',
 'dog.402.jpg',
 'dog.755.jpg',
 'dog.778.jpg',
 'dog.860.jpg',
 'dog.561.jpg',
 'dog.305.jpg',
 'dog.954.jpg',
 'dog.30.jpg',
 'dog.990.jpg',
 'dog.480.jpg',
 'dog.211.jpg',
 'dog.852.jpg',
 'dog.536.jpg',
 'dog.355.jpg',
 'dog.905.jpg',
 'dog.54.jpg',
 'dog.123.jpg',
 'dog.596.jpg',
 'dog.138.jpg',
 'dog.417.jpg',
 'dog.670.jpg',
 'dog.881.jpg',
 'dog.194.jpg',
 'dog.702.jpg',
 'dog.948.jpg',
 'dog.126.jpg',
 'dog.631.jpg',
 'dog.124.jpg',
 'dog.410.jpg',
 'dog.62.jpg',
 'dog.962.jpg',
 'dog.56.jpg',
 'dog.941.jpg',
 'dog.685.jpg',
 'dog.118.jpg',
 'dog.424.jpg',
 'dog.513.jpg',
 'dog.699.jpg',
 'dog.218.jpg',
 'dog.141.jpg',
 'dog.107.jpg',
 'dog.163.jpg',
 'dog.384.jpg',
 'dog.175.jpg',
 'dog.818.jpg',
 'dog.121.jpg',
 'dog.553.jpg',
 'dog.611.jpg',
 'dog.750.jpg',
 'dog.20.jpg',
 'dog.714.jpg',
 'dog.991.jpg',
 'dog.779.jpg',
 'dog.837.jpg',
 'dog.976.jpg',
 'dog.21.jpg',
 'dog.655.jpg',
 'dog.915.jpg',
 'dog.304.jpg',
 'dog.726.jpg',
 'dog.295.jpg',
 'dog.700.jpg',
 'dog.134.jpg',
 'dog.360.jpg',
 'dog.354.jpg',
 'dog.193.jpg',
 'dog.63.jpg',
 'dog.877.jpg',
 'dog.172.jpg',
 'dog.635.jpg',
 'dog.493.jpg',
 'dog.964.jpg',
 'dog.947.jpg',
 'dog.575.jpg',
 'dog.956.jpg',
 'dog.534.jpg',
 'dog.10.jpg',
 'dog.876.jpg',
 'dog.866.jpg',
 'dog.117.jpg',
 'dog.591.jpg',
 'dog.797.jpg',
 'dog.512.jpg',
 'dog.520.jpg',
 'dog.757.jpg',
 'dog.521.jpg',
 'dog.484.jpg',
 'dog.500.jpg',
 'dog.148.jpg',
 'dog.279.jpg',
 'dog.583.jpg',
 'dog.981.jpg',
 'dog.639.jpg',
 'dog.221.jpg',
 'dog.316.jpg',
 'dog.364.jpg',
 'dog.530.jpg',
 'dog.104.jpg',
 'dog.888.jpg',
 'dog.208.jpg',
 'dog.129.jpg',
 'dog.505.jpg',
 'dog.776.jpg',
 'dog.404.jpg',
 'dog.830.jpg',
 'dog.232.jpg',
 'dog.246.jpg',
 'dog.72.jpg',
 'dog.207.jpg',
 'dog.966.jpg',
 'dog.287.jpg',
 'dog.683.jpg',
 'dog.758.jpg',
 'dog.164.jpg',
 'dog.988.jpg',
 'dog.526.jpg',
 'dog.538.jpg',
 'dog.474.jpg',
 'dog.111.jpg',
 'dog.229.jpg',
 'dog.83.jpg',
 'dog.209.jpg',
 'dog.658.jpg',
 'dog.191.jpg',
 'dog.814.jpg',
 'dog.488.jpg',
 'dog.746.jpg',
 'dog.719.jpg',
 'dog.497.jpg',
 'dog.367.jpg',
 'dog.673.jpg',
 'dog.35.jpg',
 'dog.242.jpg',
 'dog.918.jpg',
 'dog.447.jpg',
 'dog.103.jpg',
 'dog.277.jpg',
 'dog.248.jpg',
 'dog.75.jpg',
 'dog.485.jpg',
 'dog.414.jpg',
 'dog.808.jpg',
 'dog.850.jpg',
 'dog.67.jpg',
 'dog.749.jpg',
 'dog.928.jpg',
 'dog.571.jpg',
 'dog.614.jpg',
 'dog.671.jpg',
 'dog.265.jpg',
 'dog.253.jpg',
 'dog.549.jpg',
 'dog.609.jpg',
 'dog.328.jpg',
 'dog.214.jpg',
 'dog.184.jpg',
 'dog.461.jpg',
 'dog.39.jpg',
 'dog.113.jpg',
 'dog.828.jpg',
 'dog.392.jpg',
 'dog.422.jpg',
 'dog.707.jpg',
 'dog.643.jpg',
 'dog.170.jpg',
 'dog.315.jpg',
 'dog.768.jpg',
 'dog.463.jpg',
 'dog.217.jpg',
 'dog.352.jpg',
 'dog.91.jpg',
 'dog.357.jpg',
 'dog.98.jpg',
 'dog.386.jpg',
 'dog.243.jpg',
 'dog.200.jpg',
 'dog.662.jpg',
 'dog.848.jpg',
 'dog.542.jpg',
 'dog.742.jpg',
 'dog.897.jpg',
 'dog.545.jpg',
 'dog.475.jpg',
 'dog.110.jpg',
 'dog.97.jpg',
 'dog.634.jpg',
 'dog.874.jpg',
 'dog.980.jpg',
 'dog.605.jpg',
 'dog.483.jpg',
 'dog.224.jpg',
 'dog.385.jpg',
 'dog.616.jpg',
 'dog.501.jpg',
 'dog.802.jpg',
 'dog.87.jpg',
 'dog.399.jpg',
 'dog.73.jpg',
 'dog.359.jpg',
 'dog.711.jpg',
 'dog.701.jpg',
 'dog.93.jpg',
 'dog.365.jpg',
 'dog.767.jpg',
 'dog.186.jpg',
 'dog.984.jpg',
 'dog.910.jpg',
 'dog.162.jpg',
 'dog.179.jpg',
 'dog.999.jpg',
 'dog.215.jpg',
 'dog.951.jpg',
 'dog.558.jpg',
 'dog.721.jpg',
 'dog.77.jpg',
 'dog.960.jpg',
 'dog.337.jpg',
 'dog.641.jpg',
 'dog.732.jpg',
 'dog.968.jpg',
 'dog.282.jpg',
 'dog.829.jpg',
 'dog.815.jpg',
 'dog.300.jpg',
 'dog.390.jpg',
 'dog.516.jpg',
 'dog.955.jpg',
 'dog.995.jpg',
 'dog.432.jpg',
 'dog.389.jpg',
 'dog.303.jpg',
 'dog.544.jpg',
 'dog.557.jpg',
 'dog.737.jpg',
 'dog.79.jpg',
 'dog.787.jpg',
 'dog.252.jpg',
 'dog.94.jpg',
 'dog.773.jpg',
 'dog.455.jpg',
 'dog.792.jpg',
 'dog.693.jpg',
 'dog.203.jpg',
 'dog.930.jpg',
 'dog.913.jpg',
 'dog.448.jpg',
 'dog.570.jpg',
 'dog.539.jpg',
 'dog.697.jpg',
 'dog.875.jpg',
 'dog.663.jpg',
 'dog.189.jpg',
 'dog.760.jpg',
 'dog.312.jpg',
 'dog.588.jpg',
 'dog.925.jpg',
 'dog.554.jpg',
 'dog.672.jpg',
 'dog.944.jpg',
 'dog.518.jpg',
 'dog.202.jpg',
 'dog.506.jpg',
 'dog.581.jpg',
 'dog.794.jpg',
 'dog.998.jpg',
 'dog.17.jpg',
 'dog.233.jpg',
 'dog.61.jpg',
 'dog.241.jpg',
 'dog.704.jpg',
 'dog.140.jpg',
 'dog.716.jpg',
 'dog.459.jpg',
 'dog.753.jpg',
 'dog.945.jpg',
 'dog.973.jpg',
 'dog.291.jpg',
 'dog.201.jpg',
 'dog.932.jpg',
 'dog.541.jpg',
 'dog.330.jpg',
 'dog.396.jpg',
 'dog.911.jpg',
 'dog.130.jpg',
 'dog.335.jpg',
 'dog.24.jpg',
 'dog.264.jpg',
 'dog.679.jpg',
 'dog.477.jpg',
 'dog.437.jpg',
 'dog.78.jpg',
 'dog.313.jpg',
 'dog.472.jpg',
 'dog.314.jpg',
 'dog.511.jpg',
 'dog.851.jpg',
 'dog.736.jpg',
 'dog.566.jpg',
 'dog.450.jpg',
 'dog.285.jpg',
 'dog.321.jpg',
 'dog.517.jpg',
 'dog.418.jpg',
 'dog.660.jpg',
 'dog.640.jpg',
 'dog.333.jpg',
 'dog.567.jpg',
 'dog.473.jpg',
 'dog.674.jpg',
 'dog.547.jpg',
 'dog.649.jpg',
 'dog.878.jpg',
 'dog.125.jpg',
 'dog.415.jpg',
 'dog.766.jpg',
 'dog.74.jpg',
 'dog.440.jpg',
 'dog.234.jpg',
 'dog.687.jpg',
 'dog.822.jpg',
 'dog.939.jpg',
 'dog.46.jpg',
 'dog.82.jpg',
 'dog.620.jpg',
 'dog.644.jpg',
 'dog.100.jpg',
 'dog.167.jpg',
 'dog.47.jpg',
 'dog.92.jpg',
 'dog.294.jpg',
 'dog.267.jpg',
 'dog.431.jpg',
 'dog.199.jpg',
 'dog.601.jpg',
 'dog.523.jpg',
 'dog.982.jpg',
 'dog.922.jpg',
 'dog.817.jpg',
 'dog.171.jpg',
 'dog.900.jpg',
 'dog.593.jpg',
 'dog.618.jpg',
 'dog.615.jpg',
 'dog.886.jpg',
 'dog.709.jpg',
 'dog.458.jpg',
 'dog.869.jpg',
 'dog.645.jpg',
 'dog.936.jpg',
 'dog.585.jpg',
 'dog.564.jpg',
 'dog.903.jpg',
 'dog.377.jpg',
 'dog.798.jpg',
 'dog.929.jpg',
 'dog.445.jpg',
 'dog.363.jpg',
 'dog.978.jpg',
 'dog.192.jpg',
 'dog.26.jpg',
 'dog.607.jpg',
 'dog.108.jpg',
 'dog.648.jpg',
 'dog.11.jpg',
 'dog.621.jpg',
 'dog.717.jpg',
 'dog.391.jpg',
 'dog.770.jpg',
 'dog.924.jpg',
 'dog.350.jpg',
 'dog.813.jpg',
 'dog.261.jpg',
 'dog.216.jpg',
 'dog.529.jpg',
 'dog.612.jpg',
 'dog.562.jpg',
 'dog.838.jpg',
 'dog.373.jpg',
 'dog.912.jpg',
 'dog.806.jpg',
 'dog.825.jpg',
 'dog.498.jpg',
 'dog.154.jpg',
 'dog.308.jpg',
 'dog.556.jpg',
 'dog.832.jpg',
 'dog.114.jpg',
 'dog.805.jpg',
 'dog.258.jpg',
 'dog.790.jpg',
 'dog.846.jpg',
 'dog.165.jpg',
 'dog.388.jpg',
 'dog.540.jpg',
 'dog.619.jpg',
 'dog.865.jpg',
 'dog.343.jpg',
 'dog.633.jpg',
 'dog.0.jpg',
 'dog.347.jpg',
 'dog.220.jpg',
 'dog.379.jpg',
 'dog.299.jpg',
 'dog.254.jpg',
 'dog.159.jpg',
 'dog.257.jpg',
 'dog.460.jpg',
 'dog.256.jpg',
 'dog.338.jpg',
 'dog.975.jpg',
 'dog.90.jpg',
 'dog.482.jpg',
 'dog.811.jpg',
 'dog.155.jpg',
 'dog.382.jpg',
 'dog.587.jpg',
 'dog.985.jpg',
 'dog.250.jpg',
 'dog.532.jpg',
 'dog.677.jpg',
 'dog.582.jpg',
 'dog.275.jpg',
 'dog.290.jpg',
 'dog.515.jpg',
 'dog.449.jpg',
 'dog.987.jpg',
 'dog.706.jpg',
 'dog.594.jpg',
 'dog.691.jpg',
 'dog.40.jpg',
 'dog.408.jpg',
 'dog.608.jpg',
 'dog.935.jpg',
 'dog.451.jpg',
 'dog.849.jpg',
 'dog.12.jpg',
 'dog.584.jpg',
 'dog.306.jpg',
 'dog.69.jpg',
 'dog.325.jpg',
 'dog.397.jpg',
 'dog.996.jpg',
 'dog.84.jpg',
 'dog.836.jpg',
 'dog.157.jpg',
 'dog.272.jpg',
 'dog.429.jpg',
 'dog.870.jpg',
 'dog.406.jpg',
 'dog.740.jpg',
 'dog.572.jpg',
 'dog.820.jpg',
 'dog.502.jpg',
 'dog.752.jpg',
 'dog.142.jpg',
 'dog.9.jpg',
 'dog.508.jpg',
 'dog.7.jpg',
 'dog.524.jpg',
 'dog.834.jpg',
 'dog.456.jpg',
 'dog.423.jpg',
 'dog.169.jpg',
 'dog.412.jpg',
 'dog.891.jpg',
 'dog.598.jpg',
 'dog.146.jpg',
 'dog.174.jpg',
 'dog.739.jpg',
 'dog.70.jpg',
 'dog.713.jpg',
 'dog.183.jpg',
 'dog.259.jpg',
 'dog.659.jpg',
 'dog.86.jpg',
 'dog.678.jpg',
 'dog.41.jpg',
 'dog.632.jpg',
 'dog.785.jpg',
 'dog.823.jpg',
 'dog.733.jpg',
 'dog.496.jpg',
 'dog.499.jpg',
 'dog.689.jpg',
 'dog.879.jpg',
 'dog.225.jpg',
 'dog.843.jpg',
 'dog.95.jpg',
 'dog.22.jpg',
 'dog.684.jpg',
 'dog.971.jpg',
 'dog.311.jpg',
 'dog.153.jpg',
 'dog.967.jpg',
 'dog.669.jpg',
 'dog.178.jpg',
 'dog.845.jpg',
 'dog.439.jpg',
 'dog.400.jpg',
 'dog.433.jpg',
 'dog.489.jpg',
 'dog.527.jpg',
 'dog.490.jpg',
 'dog.856.jpg',
 'dog.57.jpg',
 'dog.777.jpg',
 'dog.53.jpg',
 'dog.812.jpg',
 'dog.452.jpg',
 'dog.937.jpg',
 'dog.682.jpg',
 'dog.143.jpg',
 'dog.537.jpg',
 'dog.4.jpg',
 'dog.920.jpg',
 'dog.734.jpg',
 'dog.462.jpg',
 'dog.286.jpg',
 'dog.51.jpg',
 'dog.646.jpg',
 'dog.145.jpg',
 'dog.466.jpg',
 'dog.245.jpg',
 'dog.664.jpg',
 'dog.29.jpg',
 'dog.32.jpg',
 'dog.50.jpg',
 'dog.590.jpg',
 'dog.744.jpg',
 'dog.884.jpg',
 'dog.85.jpg',
 'dog.14.jpg',
 'dog.647.jpg',
 'dog.372.jpg',
 'dog.135.jpg',
 'dog.160.jpg',
 'dog.855.jpg',
 'dog.8.jpg',
 'dog.301.jpg',
 'dog.345.jpg',
 'dog.441.jpg',
 'dog.237.jpg',
 'dog.342.jpg',
 'dog.775.jpg',
 'dog.867.jpg',
 'dog.943.jpg',
 'dog.425.jpg',
 'dog.904.jpg',
 'dog.96.jpg',
 'dog.88.jpg',
 'dog.101.jpg',
 'dog.309.jpg',
 'dog.361.jpg',
 'dog.595.jpg',
 'dog.226.jpg',
 'dog.965.jpg',
 'dog.376.jpg',
 'dog.471.jpg',
 'dog.574.jpg',
 'dog.839.jpg',
 'dog.606.jpg',
 'dog.158.jpg',
 'dog.504.jpg',
 'dog.804.jpg',
 'dog.942.jpg',
 'dog.122.jpg',
 'dog.329.jpg',
 'dog.555.jpg',
 'dog.284.jpg',
 'dog.33.jpg',
 'dog.791.jpg',
 'dog.369.jpg',
 'dog.909.jpg',
 'dog.759.jpg',
 'dog.283.jpg',
 'dog.411.jpg',
 'dog.563.jpg',
 'dog.927.jpg',
 'dog.268.jpg',
 'dog.977.jpg',
 'dog.116.jpg',
 'dog.703.jpg',
 'dog.637.jpg',
 'dog.278.jpg',
 'dog.795.jpg',
 'dog.735.jpg',
 'dog.908.jpg',
 'dog.65.jpg',
 'dog.712.jpg',
 'dog.37.jpg',
 'dog.362.jpg',
 'dog.374.jpg',
 'dog.626.jpg',
 'dog.137.jpg',
 'dog.144.jpg',
 'dog.45.jpg',
 'dog.416.jpg',
 'dog.464.jpg',
 'dog.276.jpg',
 'dog.58.jpg',
 'dog.68.jpg',
 'dog.297.jpg',
 'dog.840.jpg',
 'dog.708.jpg',
 'dog.731.jpg',
 'dog.729.jpg',
 'dog.443.jpg',
 'dog.5.jpg',
 'dog.835.jpg',
 'dog.972.jpg',
 'dog.629.jpg',
 'dog.921.jpg',
 'dog.393.jpg',
 'dog.743.jpg',
 'dog.592.jpg',
 'dog.994.jpg',
 'dog.419.jpg',
 'dog.696.jpg',
 'dog.857.jpg',
 'dog.844.jpg',
 'dog.519.jpg',
 'dog.603.jpg',
 'dog.665.jpg',
 'dog.871.jpg',
 'dog.332.jpg',
 'dog.949.jpg',
 'dog.403.jpg',
 'dog.569.jpg',
 'dog.864.jpg',
 'dog.213.jpg',
 'dog.718.jpg',
 'dog.986.jpg',
 'dog.578.jpg',
 'dog.754.jpg',
 'dog.509.jpg',
 'dog.576.jpg',
 'dog.756.jpg',
 'dog.946.jpg',
 'dog.638.jpg',
 'dog.326.jpg',
 'dog.653.jpg',
 'dog.826.jpg',
 'dog.531.jpg',
 'dog.105.jpg',
 'dog.340.jpg',
 'dog.133.jpg',
 'dog.210.jpg',
 'dog.181.jpg',
 'dog.923.jpg',
 'dog.762.jpg',
 'dog.236.jpg',
 'dog.952.jpg',
 'dog.934.jpg',
 'dog.3.jpg',
 'dog.800.jpg',
 'dog.99.jpg',
 'dog.274.jpg',
 'dog.366.jpg',
 'dog.132.jpg',
 'dog.890.jpg',
 'dog.89.jpg',
 'dog.938.jpg',
 'dog.371.jpg',
 'dog.353.jpg',
 'dog.454.jpg',
 'dog.764.jpg',
 'dog.341.jpg',
 'dog.139.jpg',
 'dog.728.jpg',
 'dog.289.jpg',
 'dog.375.jpg',
 'dog.244.jpg',
 'dog.16.jpg',
 'dog.906.jpg',
 'dog.381.jpg',
 'dog.771.jpg',
 'dog.320.jpg',
 'dog.667.jpg',
 'dog.131.jpg',
 'dog.772.jpg',
 'dog.901.jpg',
 'dog.60.jpg',
 'dog.597.jpg',
 'dog.444.jpg',
 'dog.36.jpg',
 'dog.917.jpg',
 'dog.676.jpg',
 'dog.421.jpg',
 'dog.407.jpg',
 'dog.55.jpg',
 'dog.627.jpg',
 'dog.914.jpg',
 'dog.206.jpg',
 'dog.23.jpg',
 'dog.31.jpg',
 'dog.747.jpg',
 'dog.271.jpg',
 'dog.899.jpg',
 'dog.695.jpg',
 'dog.872.jpg',
 'dog.323.jpg',
 'dog.317.jpg',
 'dog.854.jpg',
 'dog.180.jpg',
 'dog.799.jpg',
 'dog.692.jpg',
 'dog.156.jpg',
 'dog.260.jpg',
 'dog.841.jpg',
 'dog.842.jpg',
 'dog.222.jpg',
 'dog.528.jpg',
 'dog.690.jpg',
 'dog.368.jpg',
 'dog.642.jpg',
 'dog.654.jpg',
 'dog.953.jpg',
 'dog.195.jpg',
 'dog.331.jpg',
 'dog.765.jpg',
 'dog.128.jpg',
 'dog.446.jpg',
 'dog.720.jpg',
 'dog.589.jpg',
 'dog.650.jpg',
 'dog.238.jpg',
 'dog.600.jpg',
 'dog.769.jpg',
 'dog.405.jpg',
 'dog.933.jpg',
 'dog.969.jpg',
 'dog.112.jpg',
 'dog.198.jpg',
 'dog.152.jpg',
 'dog.579.jpg',
 'dog.430.jpg',
 'dog.71.jpg',
 'dog.858.jpg',
 'dog.346.jpg',
 'dog.383.jpg',
 'dog.821.jpg',
 'dog.401.jpg',
 'dog.958.jpg',
 'dog.476.jpg',
 'dog.666.jpg',
 'dog.231.jpg',
 'dog.394.jpg',
 'dog.889.jpg',
 'dog.577.jpg',
 'dog.59.jpg',
 'dog.266.jpg',
 'dog.816.jpg',
 'dog.548.jpg',
 'dog.187.jpg',
 'dog.831.jpg',
 'dog.859.jpg',
 'dog.380.jpg',
 'dog.979.jpg',
 'dog.559.jpg',
 'dog.34.jpg',
 'dog.240.jpg',
 'dog.149.jpg',
 'dog.725.jpg',
 'dog.334.jpg',
 'dog.150.jpg',
 'dog.763.jpg',
 'dog.880.jpg',
 'dog.219.jpg',
 'dog.543.jpg',
 'dog.992.jpg',
 'dog.722.jpg',
 'dog.491.jpg',
 'dog.882.jpg',
 'dog.465.jpg',
 'dog.319.jpg',
 'dog.327.jpg',
 'dog.398.jpg',
 'dog.630.jpg',
 'dog.710.jpg',
 'dog.255.jpg',
 'dog.190.jpg',
 'dog.507.jpg',
 'dog.223.jpg',
 'dog.610.jpg',
 'dog.119.jpg',
 'dog.127.jpg',
 'dog.263.jpg',
 'dog.810.jpg',
 'dog.115.jpg',
 'dog.727.jpg',
 'dog.503.jpg',
 'dog.892.jpg',
 'dog.239.jpg',
 'dog.807.jpg',
 'dog.963.jpg',
 'dog.348.jpg',
 'dog.293.jpg',
 'dog.185.jpg',
 'dog.622.jpg',
 'dog.613.jpg',
 'dog.227.jpg',
 'dog.176.jpg',
 'dog.426.jpg',
 'dog.457.jpg',
 'dog.741.jpg',
 'dog.896.jpg',
 'dog.931.jpg',
 'dog.789.jpg',
 'dog.48.jpg',
 'dog.486.jpg',
 'dog.853.jpg',
 'dog.862.jpg',
 'dog.38.jpg',
 'dog.657.jpg',
 'dog.166.jpg',
 'dog.651.jpg',
 'dog.533.jpg',
 'dog.809.jpg',
 'dog.565.jpg',
 'dog.2.jpg',
 'dog.66.jpg',
 'dog.177.jpg',
 'dog.916.jpg',
 'dog.551.jpg',
 'dog.228.jpg',
 'dog.182.jpg',
 'dog.730.jpg',
 'dog.310.jpg',
 'dog.788.jpg',
 'dog.705.jpg',
 'dog.604.jpg',
 'dog.688.jpg',
 'dog.336.jpg',
 'dog.681.jpg',
 'dog.6.jpg',
 'dog.136.jpg',
 'dog.196.jpg',
 'dog.168.jpg',
 'dog.628.jpg',
 'dog.212.jpg',
 'dog.351.jpg',
 'dog.950.jpg',
 'dog.885.jpg',
 'dog.573.jpg',
 'dog.281.jpg',
 'dog.959.jpg',
 'dog.467.jpg',
 'dog.109.jpg',
 'dog.470.jpg',
 'dog.668.jpg',
 'dog.1.jpg',
 'dog.120.jpg',
 'dog.887.jpg',
 'dog.546.jpg',
 'dog.898.jpg',
 'dog.479.jpg',
 'dog.656.jpg',
 'dog.624.jpg',
 'dog.318.jpg',
 'dog.173.jpg',
 'dog.782.jpg',
 'dog.49.jpg',
 'dog.919.jpg',
 'dog.599.jpg',
 'dog.873.jpg',
 'dog.356.jpg',
 'dog.957.jpg',
 'dog.617.jpg',
 'dog.997.jpg',
 'dog.989.jpg',
 'dog.262.jpg',
 'dog.781.jpg',
 'dog.494.jpg',
 'dog.15.jpg',
 'dog.623.jpg',
 'dog.535.jpg',
 'dog.680.jpg',
 'dog.27.jpg',
 'dog.161.jpg',
 'dog.774.jpg',
 'dog.625.jpg',
 'dog.442.jpg',
 'dog.28.jpg',
 'dog.661.jpg',
 'dog.636.jpg',
 'dog.44.jpg',
 'dog.302.jpg',
 'dog.793.jpg',
 'dog.780.jpg',
 'dog.983.jpg',
 'dog.510.jpg',
 'dog.861.jpg',
 'dog.324.jpg',
 'dog.64.jpg',
 'dog.188.jpg',
 'dog.487.jpg',
 'dog.940.jpg',
 'dog.652.jpg',
 'dog.249.jpg',
 'dog.894.jpg',
 'dog.803.jpg',
 'dog.102.jpg',
 'dog.970.jpg',
 'dog.761.jpg',
 'dog.378.jpg',
 'dog.307.jpg',
 'dog.358.jpg',
 'dog.751.jpg',
 'dog.738.jpg']

In [ ]:

In [ ]:

트레이닝 이미지와 밸리데이션 이미지를, 각각 몇개씩인지 확인해 본다.¶

In [12]:

len(os.listdir(train_dir+'/dogs'))

Out[12]:

1000

In [13]:

len(os.listdir(train_dir+'/cats'))

Out[13]:

1000

In [14]:

len(os.listdir(test_dir+'/dogs'))

Out[14]:

500

In [15]:

len(os.listdir(test_dir+'/cats'))

Out[15]:

500

In [ ]:

In [ ]:

In [ ]:

Building a Small Model from Scratch to Get to ~72% Accuracy¶

이미지의 사이즈를 150x150, 칼라(rgb) 로 처리하자.

In [17]:

import tensorflow as tf
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense

In [18]:

def build_model() :
  model = Sequential()
  model.add( Conv2D(16, (3,3), activation='relu', input_shape=(150, 150, 3) ) )
  model.add( MaxPooling2D((2,2) , 2 )) 
  model.add( Conv2D(32, (3,3), activation='relu' ) )
  model.add( MaxPooling2D((2,2) , 2 ))
  model.add( Conv2D(64, (3,3), activation='relu' ) )
  model.add( MaxPooling2D((2,2) , 2 ))

  model.add(Flatten())
  model.add(Dense(512, 'relu'))
  model.add(Dense(1, 'sigmoid'))

  model.compile('rmsprop', 'binary_crossentropy', metrics=['accuracy'])
  return model

In [19]:

model = build_model()

서머리 해보자

In [20]:

model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d (Conv2D)             (None, 148, 148, 16)      448       
                                                                 
 max_pooling2d (MaxPooling2D  (None, 74, 74, 16)       0         
 )                                                               
                                                                 
 conv2d_1 (Conv2D)           (None, 72, 72, 32)        4640      
                                                                 
 max_pooling2d_1 (MaxPooling  (None, 36, 36, 32)       0         
 2D)                                                             
                                                                 
 conv2d_2 (Conv2D)           (None, 34, 34, 64)        18496     
                                                                 
 max_pooling2d_2 (MaxPooling  (None, 17, 17, 64)       0         
 2D)                                                             
                                                                 
 flatten (Flatten)           (None, 18496)             0         
                                                                 
 dense (Dense)               (None, 512)               9470464   
                                                                 
 dense_1 (Dense)             (None, 1)                 513       
                                                                 
=================================================================
Total params: 9,494,561
Trainable params: 9,494,561
Non-trainable params: 0
_________________________________________________________________

In [ ]:

In [ ]:

RMSprop 으로 컴파일한다.¶

In [ ]:

In [ ]:

Data Preprocessing¶

ImageDataGenerator 사용하기

In [21]:

from keras.preprocessing.image import ImageDataGenerator

In [24]:

train_datagen = ImageDataGenerator(rescale=1/255.0) # 폴더마다 각각 만든다

In [25]:

test_datagen = ImageDataGenerator(rescale=1/255.0)

In [26]:

train_generator = train_datagen.flow_from_directory(train_dir, target_size = (150,150), class_mode = 'binary', batch_size = 20)  # batch_size =20 == 20개씩 묶음으로 처리해라

Found 2000 images belonging to 2 classes.

In [27]:

test_generator = test_datagen.flow_from_directory(test_dir, target_size= (150,150), class_mode = 'binary', batch_size=20)

Found 1000 images belonging to 2 classes.

Training¶

15 epochs 로 학습해 보자.

In [28]:

epoch_history = model.fit(train_generator, epochs= 15, validation_data= (test_generator) , steps_per_epoch= 100 ) # steps_per_epoch=100 == batch_size=20 (아래공식에 따라, 같은것)
# 학습데이터 2000개 == 배치사이즈 X 스텝

Epoch 1/15
100/100 [==============================] - 18s 94ms/step - loss: 0.8415 - accuracy: 0.5570 - val_loss: 0.6607 - val_accuracy: 0.5960
Epoch 2/15
100/100 [==============================] - 8s 83ms/step - loss: 0.6524 - accuracy: 0.6425 - val_loss: 0.6101 - val_accuracy: 0.6970
Epoch 3/15
100/100 [==============================] - 9s 85ms/step - loss: 0.5600 - accuracy: 0.7245 - val_loss: 0.5866 - val_accuracy: 0.7070
Epoch 4/15
100/100 [==============================] - 9s 85ms/step - loss: 0.4876 - accuracy: 0.7685 - val_loss: 0.5739 - val_accuracy: 0.7320
Epoch 5/15
100/100 [==============================] - 9s 86ms/step - loss: 0.4070 - accuracy: 0.8085 - val_loss: 0.6023 - val_accuracy: 0.7130
Epoch 6/15
100/100 [==============================] - 8s 84ms/step - loss: 0.3241 - accuracy: 0.8645 - val_loss: 0.6906 - val_accuracy: 0.7140
Epoch 7/15
100/100 [==============================] - 9s 86ms/step - loss: 0.2630 - accuracy: 0.8935 - val_loss: 0.6666 - val_accuracy: 0.7310
Epoch 8/15
100/100 [==============================] - 10s 98ms/step - loss: 0.1858 - accuracy: 0.9265 - val_loss: 0.7943 - val_accuracy: 0.7250
Epoch 9/15
100/100 [==============================] - 9s 90ms/step - loss: 0.1207 - accuracy: 0.9560 - val_loss: 0.7994 - val_accuracy: 0.7020
Epoch 10/15
100/100 [==============================] - 8s 85ms/step - loss: 0.0887 - accuracy: 0.9740 - val_loss: 1.1665 - val_accuracy: 0.7090
Epoch 11/15
100/100 [==============================] - 8s 85ms/step - loss: 0.0731 - accuracy: 0.9710 - val_loss: 1.1989 - val_accuracy: 0.7160
Epoch 12/15
100/100 [==============================] - 9s 87ms/step - loss: 0.0608 - accuracy: 0.9795 - val_loss: 1.3441 - val_accuracy: 0.7200
Epoch 13/15
100/100 [==============================] - 9s 86ms/step - loss: 0.0585 - accuracy: 0.9845 - val_loss: 1.5738 - val_accuracy: 0.7210
Epoch 14/15
100/100 [==============================] - 8s 85ms/step - loss: 0.0459 - accuracy: 0.9865 - val_loss: 2.3166 - val_accuracy: 0.6600
Epoch 15/15
100/100 [==============================] - 9s 85ms/step - loss: 0.0196 - accuracy: 0.9930 - val_loss: 2.0281 - val_accuracy: 0.7160

In [29]:

# 평가
model.evaluate(test_generator)

50/50 [==============================] - 3s 56ms/step - loss: 2.0281 - accuracy: 0.7160

Out[29]:

[2.028092384338379, 0.7160000205039978]

In [ ]:

Running the Model¶

픽사베이에서 무료 사진 다운로드 받아서, 실제로 예측해 보자

In [30]:

# 잘돌아가는지 테스트
import numpy as np
from google.colab import files
from tensorflow.keras.preprocessing import image

uploaded = files.upload()

for fn in uploaded.keys() :
  path = '/content/' + fn             # 콘텍트디렉토리 저장
  img = image.load_img(path, target_size=(150,150))  
  x = image.img_to_array(img) / 255.0  
  print(x)
  print(x.shape)

  x = np.expand_dims(x, axis = 0) # 차원변경

  print(x.shape)

  images = np.vstack( [x] )
  classes = model.predict( images, batch_size = 10 )
  
  print(classes)

  if classes[0] > 0.5 :              # 10개분류면 if가 10개 
    print(fn + " is a dog")
  else :
    print(fn + " is a cat")

Upload widget is only available when the cell has been executed in the current browser session. Please rerun this cell to enable.

Saving bbb.jpg to bbb.jpg
[[[0.84313726 0.8156863  0.7764706 ]
  [0.8392157  0.8117647  0.77254903]
  [0.84705883 0.81960785 0.78039217]
  ...
  [0.6862745  0.6901961  0.6666667 ]
  [0.6901961  0.69411767 0.67058825]
  [0.7019608  0.7058824  0.6745098 ]]

 [[0.84313726 0.8156863  0.7764706 ]
  [0.84705883 0.81960785 0.78039217]
  [0.84705883 0.81960785 0.78039217]
  ...
  [0.6901961  0.69411767 0.67058825]
  [0.69803923 0.7019608  0.6784314 ]
  [0.7058824  0.70980394 0.6862745 ]]

 [[0.84313726 0.81960785 0.77254903]
  [0.84705883 0.8235294  0.7764706 ]
  [0.84705883 0.81960785 0.78039217]
  ...
  [0.69411767 0.69803923 0.6745098 ]
  [0.7019608  0.7058824  0.68235296]
  [0.7058824  0.70980394 0.6862745 ]]

 ...

 [[0.7529412  0.7372549  0.69411767]
  [0.7607843  0.74509805 0.7019608 ]
  [0.76862746 0.7529412  0.70980394]
  ...
  [0.654902   0.6431373  0.6862745 ]
  [0.6431373  0.627451   0.68235296]
  [0.627451   0.6117647  0.67058825]]

 [[0.7529412  0.7372549  0.69411767]
  [0.7607843  0.74509805 0.7019608 ]
  [0.7647059  0.7490196  0.7058824 ]
  ...
  [0.6509804  0.6392157  0.68235296]
  [0.6431373  0.627451   0.68235296]
  [0.63529414 0.61960787 0.6745098 ]]

 [[0.7490196  0.73333335 0.6901961 ]
  [0.7607843  0.74509805 0.7019608 ]
  [0.7647059  0.7490196  0.7058824 ]
  ...
  [0.654902   0.6313726  0.6784314 ]
  [0.6431373  0.6313726  0.6745098 ]
  [0.6392157  0.6156863  0.6627451 ]]]
(150, 150, 3)
(1, 150, 150, 3)
1/1 [==============================] - 0s 175ms/step
[[0.99999857]]
bbb.jpg is a dog

In [ ]:

Evaluating Accuracy and Loss for the Model¶

training/validation accuracy 와 loss 를 차트로 시각화 한다.

In [ ]:

In [ ]:

In [ ]:

overfitting 을 확인해 보자