This project is part of deeplearning nanodegree on udacity implemented using pytorch
Welcome to the Convolutional Neural Networks (CNN) project in the AI Nanodegree! In this project, you will learn how to build a pipeline that can be used within a web or mobile app to process real-world, user-supplied images. Given an image of a dog, your algorithm will identify an estimate of the canine’s breed. If supplied an image of a human, the code will identify the resembling dog breed.
Along with exploring state-of-the-art CNN models for classification and localization, you will make important design decisions about the user experience for your app. Our goal is that by completing this lab, you understand the challenges involved in piecing together a series of models designed to perform various tasks in a data processing pipeline. Each model has its strengths and weaknesses, and engineering a real-world application often involves solving many problems without a perfect answer. Your imperfect solution will nonetheless create a fun user experience!
- Download the dog dataset.
- Download the human dataset.
CNN Structures (Building a model on my own) (conv1): Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
activation: relu
(pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
activation: relu
(conv2): Conv2d(32, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
activation: relu
(dropout): Dropout(p=0.3)
(pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(conv3): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(conv4): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(dropout): Dropout(p=0.3)
(fc1): Linear(in_features=256*3*3, out_features=512, bias=True)
(dropout): Dropout(p=0.3)
(fc2): Linear(in_features=512, out_features=133, bias=True)
Used ResNet50 for transfer learning by changing last layer of pretrained network with adam optimizer at learning rate of 0.001
Epoch 1, Batch 1 loss: 4.988276
Epoch 1, Batch 101 loss: 3.990508
Epoch 1, Batch 201 loss: 3.050565
Epoch 1, Batch 301 loss: 2.610464
Epoch: 1 Training Loss: 2.497172 Validation Loss: 0.806791
Validation loss decreased (inf --> 0.806791). Saving model ...
Epoch 2, Batch 1 loss: 1.671902
Epoch 2, Batch 101 loss: 1.302282
Epoch 2, Batch 201 loss: 1.275751
Epoch 2, Batch 301 loss: 1.276304
Epoch: 2 Training Loss: 1.271454 Validation Loss: 0.589309
Validation loss decreased (0.806791 --> 0.589309). Saving model ...
Epoch 3, Batch 1 loss: 0.628817
Epoch 3, Batch 101 loss: 1.083294
Epoch 3, Batch 201 loss: 1.088256
Epoch 3, Batch 301 loss: 1.097040
Epoch: 3 Training Loss: 1.083551 Validation Loss: 0.548926
Validation loss decreased (0.589309 --> 0.548926). Saving model ...
Epoch 4, Batch 1 loss: 1.490823
Epoch 4, Batch 101 loss: 0.999596
Epoch 4, Batch 201 loss: 0.997450
Epoch 4, Batch 301 loss: 0.997139
Epoch: 4 Training Loss: 0.997130 Validation Loss: 0.481983
Validation loss decreased (0.548926 --> 0.481983). Saving model ...
Epoch 5, Batch 1 loss: 0.673107
Epoch 5, Batch 101 loss: 0.939902
Epoch 5, Batch 201 loss: 0.933506
Epoch 5, Batch 301 loss: 0.954241
Epoch: 5 Training Loss: 0.956093 Validation Loss: 0.477505
Validation loss decreased (0.481983 --> 0.477505). Saving model ...
Epoch 6, Batch 1 loss: 0.581581
Epoch 6, Batch 101 loss: 0.871107
Epoch 6, Batch 201 loss: 0.862837
Epoch 6, Batch 301 loss: 0.884670
Epoch: 6 Training Loss: 0.884058 Validation Loss: 0.467822
Validation loss decreased (0.477505 --> 0.467822). Saving model ...
Epoch 7, Batch 1 loss: 1.427830
Epoch 7, Batch 101 loss: 0.905297
Epoch 7, Batch 201 loss: 0.910696
Epoch 7, Batch 301 loss: 0.913614
Epoch: 7 Training Loss: 0.906176 Validation Loss: 0.528639
Epoch 8, Batch 1 loss: 0.839843
Epoch 8, Batch 101 loss: 0.880858
Epoch 8, Batch 201 loss: 0.844327
Epoch 8, Batch 301 loss: 0.857063
Epoch: 8 Training Loss: 0.860199 Validation Loss: 0.530373
Epoch 9, Batch 1 loss: 0.713667
Epoch 9, Batch 101 loss: 0.873187
Epoch 9, Batch 201 loss: 0.852546
Epoch 9, Batch 301 loss: 0.875552
Epoch: 9 Training Loss: 0.882663 Validation Loss: 0.610281
Epoch 10, Batch 1 loss: 1.135839
Epoch 10, Batch 101 loss: 0.850564
Epoch 10, Batch 201 loss: 0.843890
Epoch 10, Batch 301 loss: 0.843047
Epoch: 10 Training Loss: 0.859217 Validation Loss: 0.505544
Epoch 11, Batch 1 loss: 0.586802
Epoch 11, Batch 101 loss: 0.786204
Epoch 11, Batch 201 loss: 0.806350
Epoch 11, Batch 301 loss: 0.808824
Epoch: 11 Training Loss: 0.812037 Validation Loss: 0.462997
Validation loss decreased (0.467822 --> 0.462997). Saving model ...
Epoch 12, Batch 1 loss: 0.624215
Epoch 12, Batch 101 loss: 0.786815
Epoch 12, Batch 201 loss: 0.831431
Epoch 12, Batch 301 loss: 0.842150
Epoch: 12 Training Loss: 0.837809 Validation Loss: 0.461257
Validation loss decreased (0.462997 --> 0.461257). Saving model ...
Epoch 13, Batch 1 loss: 0.245320
Epoch 13, Batch 101 loss: 0.774077
Epoch 13, Batch 201 loss: 0.781466
Epoch 13, Batch 301 loss: 0.795114
Epoch: 13 Training Loss: 0.799339 Validation Loss: 0.519438
Epoch 14, Batch 1 loss: 0.350892
Epoch 14, Batch 101 loss: 0.755855
Epoch 14, Batch 201 loss: 0.795633
Epoch 14, Batch 301 loss: 0.802520
Epoch: 14 Training Loss: 0.801363 Validation Loss: 0.479285
Epoch 15, Batch 1 loss: 1.615883
Epoch 15, Batch 101 loss: 0.701669
Epoch 15, Batch 201 loss: 0.729113
Epoch 15, Batch 301 loss: 0.755251
Epoch: 15 Training Loss: 0.774621 Validation Loss: 0.549827
Epoch 16, Batch 1 loss: 0.560869
Epoch 16, Batch 101 loss: 0.814576
Epoch 16, Batch 201 loss: 0.800165
Epoch 16, Batch 301 loss: 0.792618
Epoch: 16 Training Loss: 0.800380 Validation Loss: 0.543387
Epoch 17, Batch 1 loss: 0.142199
Epoch 17, Batch 101 loss: 0.780947
Epoch 17, Batch 201 loss: 0.751923
Epoch 17, Batch 301 loss: 0.768835
Epoch: 17 Training Loss: 0.769857 Validation Loss: 0.481146
Epoch 18, Batch 1 loss: 0.449748
Epoch 18, Batch 101 loss: 0.820581
Epoch 18, Batch 201 loss: 0.782869
Epoch 18, Batch 301 loss: 0.791180
Epoch: 18 Training Loss: 0.780486 Validation Loss: 0.556208
Epoch 19, Batch 1 loss: 1.166562
Epoch 19, Batch 101 loss: 0.820666
Epoch 19, Batch 201 loss: 0.812832
Epoch 19, Batch 301 loss: 0.784900
Epoch: 19 Training Loss: 0.788714 Validation Loss: 0.521605
Epoch 20, Batch 1 loss: 0.761048
Epoch 20, Batch 101 loss: 0.699008
Epoch 20, Batch 201 loss: 0.739878
Epoch 20, Batch 301 loss: 0.759585
Epoch: 20 Training Loss: 0.769439 Validation Loss: 0.587037
Epoch 21, Batch 1 loss: 0.935080
Epoch 21, Batch 101 loss: 0.726010
Epoch 21, Batch 201 loss: 0.771881
Epoch 21, Batch 301 loss: 0.765983
Epoch: 21 Training Loss: 0.770885 Validation Loss: 0.541668
Epoch 22, Batch 1 loss: 0.555262
Epoch 22, Batch 101 loss: 0.728653
Epoch 22, Batch 201 loss: 0.733303
Epoch 22, Batch 301 loss: 0.745302
Epoch: 22 Training Loss: 0.742251 Validation Loss: 0.527759
Epoch 23, Batch 1 loss: 0.294313
Epoch 23, Batch 101 loss: 0.735635
Epoch 23, Batch 201 loss: 0.730064
Epoch 23, Batch 301 loss: 0.758786
Epoch: 23 Training Loss: 0.749529 Validation Loss: 0.520503
Epoch 24, Batch 1 loss: 0.131095
Epoch 24, Batch 101 loss: 0.788072
Epoch 24, Batch 201 loss: 0.754351
Epoch 24, Batch 301 loss: 0.754937
Epoch: 24 Training Loss: 0.759802 Validation Loss: 0.530764
Epoch 25, Batch 1 loss: 1.208522
Epoch 25, Batch 101 loss: 0.643047
Epoch 25, Batch 201 loss: 0.705212
Epoch 25, Batch 301 loss: 0.728429
Epoch: 25 Training Loss: 0.726061 Validation Loss: 0.574407