Image Classification Using AutoNN
AutoNN's CNN (Convolutional Neural Network) Models are built on PyTorch.
Use AutoNN to classify cifar-10 Dataset
1. How to train a CNN model for image classification:
Import CreateCNN from CNN's cnn_generator module
Example
from AutoNN.CNN.cnn_generator import CreateCNN, CNN
inst = CreateCNN()
model,model_config,history=inst.get_bestCNN(path_trainset="E:/output/cifar10/cifar10/train",
path_testset="E:/output/cifar10/cifar10/test",
split_required=False,EPOCHS=10,image_shape=(32,32))
Output
░█▀▀█ █░░█ ▀▀█▀▀ █▀▀█ ▒█▄░▒█ ▒█▄░▒█
▒█▄▄█ █░░█ ░░█░░ █░░█ ▒█▒█▒█ ▒█▒█▒█
▒█░▒█ ░▀▀▀ ░░▀░░ ▀▀▀▀ ▒█░░▀█ ▒█░░▀█
Version: 2.0.0
An AutoML framework by
Anish Konar, Arjun Ghosh, Rajarshi Banerjee, Sagnik Nayak.
Default computing platform: cuda
Classes: ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck'] # Classes: 10
Training set size: 40000 | Validation Set size: 10000 | Test Set size: 10000
Architecture search Complete..! Time Taken: 0:00:01.688582
Number of models generated: 2
Searching for the best model. Please be patient. Thank you....
Training CNN model cnn0
Epoch: 1/10
100%|██████████| 2500/2500 [01:08<00:00, 36.24it/s]
Training Accuracy: 35.1775 Training Loss:1.7207594780921935
100%|██████████| 625/625 [00:09<00:00, 65.11it/s]
Validation Accuracy: 43.38 Validation Loss:1.5083902006149292
Epoch: 2/10
100%|██████████| 2500/2500 [01:03<00:00, 39.34it/s]
Training Accuracy: 44.7975 Training Loss:1.483097927236557
100%|██████████| 625/625 [00:08<00:00, 70.63it/s]
Validation Accuracy: 47.59 Validation Loss:1.4107291974067688
Epoch: 3/10
100%|██████████| 2500/2500 [01:03<00:00, 39.67it/s]
Training Accuracy: 49.26 Training Loss:1.3825817276716232
100%|██████████| 625/625 [00:08<00:00, 70.44it/s]
Validation Accuracy: 49.26 Validation Loss:1.3538662633895875
Epoch: 4/10
100%|██████████| 2500/2500 [01:02<00:00, 39.78it/s]
Training Accuracy: 51.4975 Training Loss:1.3264493201732634
100%|██████████| 625/625 [00:08<00:00, 70.79it/s]
Validation Accuracy: 54.59 Validation Loss:1.263453982925415
Epoch: 5/10
100%|██████████| 2500/2500 [01:03<00:00, 39.61it/s]
Training Accuracy: 53.3775 Training Loss:1.2788944167137146
100%|██████████| 625/625 [00:08<00:00, 70.93it/s]
Validation Accuracy: 53.4 Validation Loss:1.2734063954353332
Epoch: 6/10
100%|██████████| 2500/2500 [01:03<00:00, 39.67it/s]
Training Accuracy: 54.59 Training Loss:1.2445036834478378
100%|██████████| 625/625 [00:08<00:00, 70.81it/s]
Validation Accuracy: 54.01 Validation Loss:1.2412574873924256
Epoch: 7/10
100%|██████████| 2500/2500 [01:03<00:00, 39.66it/s]
Training Accuracy: 56.045 Training Loss:1.2121670446991921
100%|██████████| 625/625 [00:08<00:00, 70.66it/s]
Validation Accuracy: 57.49 Validation Loss:1.1828145512580872
Epoch: 8/10
100%|██████████| 2500/2500 [01:02<00:00, 39.71it/s]
Training Accuracy: 56.75 Training Loss:1.1846893740534783
100%|██████████| 625/625 [00:08<00:00, 70.39it/s]
Validation Accuracy: 58.49 Validation Loss:1.1632665138721465
Epoch: 9/10
100%|██████████| 2500/2500 [01:02<00:00, 39.73it/s]
Training Accuracy: 58.13 Training Loss:1.153260654783249
100%|██████████| 625/625 [00:08<00:00, 70.20it/s]
Validation Accuracy: 60.65 Validation Loss:1.1054361756324769
Epoch: 10/10
100%|██████████| 2500/2500 [01:02<00:00, 40.28it/s]
Training Accuracy: 58.895 Training Loss:1.1318354423761368
100%|██████████| 625/625 [00:08<00:00, 73.48it/s]
Validation Accuracy: 59.0 Validation Loss:1.1538037222385407
Calculating test accuracy CNN model cnn0
Test ACCuracy: 59.84 Test Loss: 1.13254158577919
------------------------------------------------------------------------------------------------------------------------------------------------------
______________________________________________________________________________________________________________________________________________________
Training CNN model cnn1
Epoch: 1/10
100%|██████████| 2500/2500 [00:41<00:00, 59.80it/s]
Training Accuracy: 30.5625 Training Loss:1.8546679210186006
100%|██████████| 625/625 [00:08<00:00, 76.54it/s]
Validation Accuracy: 33.79 Validation Loss:1.7818523860931397
Epoch: 2/10
100%|██████████| 2500/2500 [00:41<00:00, 60.42it/s]
Training Accuracy: 39.33 Training Loss:1.6098018644332885
100%|██████████| 625/625 [00:08<00:00, 76.10it/s]
Validation Accuracy: 43.24 Validation Loss:1.520786734867096
Epoch: 3/10
100%|██████████| 2500/2500 [00:41<00:00, 60.35it/s]
Training Accuracy: 44.2325 Training Loss:1.5038440037250518
100%|██████████| 625/625 [00:08<00:00, 76.69it/s]
Validation Accuracy: 45.07 Validation Loss:1.4924028639793396
Epoch: 4/10
100%|██████████| 2500/2500 [00:41<00:00, 60.33it/s]
Training Accuracy: 47.06 Training Loss:1.438860204720497
100%|██████████| 625/625 [00:08<00:00, 76.10it/s]
Validation Accuracy: 47.9 Validation Loss:1.4355408569335937
Epoch: 5/10
100%|██████████| 2500/2500 [00:41<00:00, 60.41it/s]
Training Accuracy: 49.3625 Training Loss:1.3855291500329971
100%|██████████| 625/625 [00:08<00:00, 76.00it/s]
Validation Accuracy: 51.42 Validation Loss:1.328598715877533
Epoch: 6/10
100%|██████████| 2500/2500 [00:41<00:00, 59.96it/s]
Training Accuracy: 51.2 Training Loss:1.3459105017662047
100%|██████████| 625/625 [00:09<00:00, 66.04it/s]
Validation Accuracy: 53.64 Validation Loss:1.2791677060127258
Epoch: 7/10
100%|██████████| 2500/2500 [00:44<00:00, 56.65it/s]
Training Accuracy: 52.4025 Training Loss:1.3105635814905168
100%|██████████| 625/625 [00:08<00:00, 69.72it/s]
Validation Accuracy: 52.93 Validation Loss:1.2803085575103759
Epoch: 8/10
100%|██████████| 2500/2500 [00:44<00:00, 56.75it/s]
Training Accuracy: 53.41 Training Loss:1.2841510282278061
100%|██████████| 625/625 [00:09<00:00, 69.40it/s]
Validation Accuracy: 55.3 Validation Loss:1.2356650713443755
Epoch: 9/10
100%|██████████| 2500/2500 [00:43<00:00, 57.20it/s]
Training Accuracy: 54.435 Training Loss:1.2552653106451035
100%|██████████| 625/625 [00:09<00:00, 67.85it/s]
Validation Accuracy: 57.19 Validation Loss:1.1992870372772217
Epoch: 10/10
100%|██████████| 2500/2500 [00:43<00:00, 57.08it/s]
Training Accuracy: 55.0625 Training Loss:1.2330288346767426
100%|██████████| 625/625 [00:09<00:00, 67.08it/s]
Validation Accuracy: 56.51 Validation Loss:1.2072079391002655
Calculating test accuracy CNN model cnn1
Test ACCuracy: 55.25 Test Loss: 1.2267251291751862
------------------------------------------------------------------------------------------------------------------------------------------------------
______________________________________________________________________________________________________________________________________________________
Best test accuracy achieved by model cnn0: 59.84
Print the model:
Output
CNN(
(network): Sequential(
(0): SkipLayer(
(skiplayers): Sequential(
(0): Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): Conv2d(16, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(skip_connection): Conv2d(3, 16, kernel_size=(1, 1), stride=(1, 1))
(relu): ReLU()
)
(1): SkipLayer(
(skiplayers): Sequential(
(0): Conv2d(16, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): Conv2d(16, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(skip_connection): Conv2d(16, 128, kernel_size=(1, 1), stride=(1, 1))
(relu): ReLU()
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(classifier): Sequential(
(0): Linear(in_features=128, out_features=32, bias=True)
(1): ReLU()
(2): Linear(in_features=32, out_features=10, bias=True)
)
)
Get list of all classes. Example:
Print Model configuration:
Print history of all training data:
Output
{
'cnn0': {
'trainloss': [1.7207594780921935, 1.483097927236557, 1.3825817276716232, 1.3264493201732634, 1.2788944167137146, 1.2445036834478378, 1.2121670446991921, 1.1846893740534783, 1.153260654783249,
1.1318354423761368],
'trainacc': [35.1775, 44.7975, 49.26, 51.4975, 53.3775, 54.59, 56.045, 56.75, 58.13, 58.895],
'valloss': [1.5083902006149292, 1.4107291974067688, 1.3538662633895875, 1.263453982925415, 1.2734063954353332, 1.2412574873924256, 1.1828145512580872, 1.1632665138721465, 1.1054361756324769, 1.1538037222385407],
'valacc': [43.38, 47.59, 49.26, 54.59, 53.4, 54.01, 57.49, 58.49, 60.65, 59.0]},
'cnn1': {
'trainloss': [1.8546679210186006, 1.6098018644332885, 1.5038440037250518, 1.438860204720497, 1.3855291500329971, 1.3459105017662047, 1.3105635814905168, 1.2841510282278061, 1.2552653106451035, 1.2330288346767426],
'trainacc': [30.5625, 39.33, 44.2325, 47.06, 49.3625, 51.2, 52.4025, 53.41, 54.435, 55.0625],
'valloss': [1.7818523860931397, 1.520786734867096, 1.4924028639793396, 1.4355408569335937, 1.328598715877533, 1.2791677060127258, 1.2803085575103759, 1.2356650713443755, 1.1992870372772217, 1.2072079391002655],
'valacc': [33.79, 43.24, 45.07, 47.9, 51.42, 53.64, 52.93, 55.3, 57.19, 56.51]
}
}
2. Plot the Training loss vs Validation loss:
3. How to save the model:
4. To check the summary of the model:
This function returns a tuple of(trainable parameters, total parameters, non-trainable parameters). See the output section below.
Output
Layer Output Shape Kernal Shape #params #(weights + bias) requires_grad
------------------------------------------------------------------------------------------------------------------------------------------------------
Conv2d-1 [1, 16, 32, 32] [16, 3, 3, 3] 448 (432 + 16) True True
BatchNorm2d-2 [1, 16, 32, 32] [16] 32 (16 + 16) True True
ReLU-3 [1, 16, 32, 32]
Conv2d-4 [1, 16, 32, 32] [16, 16, 3, 3] 2320 (2304 + 16) True True
BatchNorm2d-5 [1, 16, 32, 32] [16] 32 (16 + 16) True True
Conv2d-6 [1, 16, 32, 32] [16, 3, 1, 1] 64 (48 + 16) True True
ReLU-7 [1, 16, 32, 32]
Conv2d-8 [1, 16, 32, 32] [16, 16, 3, 3] 2320 (2304 + 16) True True
BatchNorm2d-9 [1, 16, 32, 32] [16] 32 (16 + 16) True True
ReLU-10 [1, 16, 32, 32]
Conv2d-11 [1, 128, 32, 32] [128, 16, 3, 3] 18560 (18432 + 128) True True
BatchNorm2d-12 [1, 128, 32, 32] [128] 256 (128 + 128) True True
Conv2d-13 [1, 128, 32, 32] [128, 16, 1, 1] 2176 (2048 + 128) True True
ReLU-14 [1, 128, 32, 32]
AdaptiveAvgPool2d-15 [1, 128, 1, 1]
Linear-16 [1, 32] [32, 128] 4128 (4096 + 32) True True
ReLU-17 [1, 32]
Linear-18 [1, 10] [10, 32] 330 (320 + 10) True True
______________________________________________________________________________________________________________________________________________________
Total parameters 30,698
Total Non-Trainable parameters 0
Total Trainable parameters 30,698
(30698, 30698, 0)
5. How to load saved model:
myModel = CNN(3,10)
myModel.load(PATH='./best models/mnistmodel.pth',config_path="./best models/mnistcfg.json",printmodel=True)
Output
Network Architecture loaded!
CNN(
(network): Sequential(
(0): SkipLayer(
(skiplayers): Sequential(
(0): Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): Conv2d(16, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(skip_connection): Conv2d(3, 16, kernel_size=(1, 1), stride=(1, 1))
(relu): ReLU()
)
(1): SkipLayer(
(skiplayers): Sequential(
(0): Conv2d(16, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): Conv2d(16, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(skip_connection): Conv2d(16, 128, kernel_size=(1, 1), stride=(1, 1))
(relu): ReLU()
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(classifier): Sequential(
(0): Linear(in_features=128, out_features=32, bias=True)
(1): ReLU()
(2): Linear(in_features=32, out_features=10, bias=True)
)
)
Loading complete, your model is now ready for evaluation!
6. How to use Loaded model on new images:
test_images = ['E:/output/cifar10/cifar10/test/bird/0012.png', # bird
'E:/output/cifar10/cifar10/train/ship/0002.png'] # ship
myModel.predict(paths=test_images)