A collection of projects in pytorch.
The projects are used as practice or base for further research.
Create a virtual environment.
"""
Linux
python3 -m venv env
source env/bin/activate
Windows python -m venv env env/Script/activate.bin """
Install all the third party libraries (Linux, on Windows use pip).
pytorch
pip3 install opencv-python
pip3 install torchsummary
pip3 install tensorboard
pip3 install tensorboardX
pip3 install scikit-image
pip3 install tqdm
Download the dataset DeepCrack:
https://github.com/yhlleo/DeepCrack
In a terminal, start virtual environment.
"""
tensorboard --logdir experiments
"""
Open two terminals, and for each terminal:
In a terminal:
start virtual environment
source env/bin/activate
run tensorboard
tensorboard --logdir experiments
http://localhost:6006/
In another terminal:
start virtual environment
source env/bin/activate
run samples
TODO:
[x] Connect repetitive blocks
[x] Connect different blocks
[] Custom loss
[] Custom node
[] Network from configuration file
[] Test transform ONNX an test OpenVINO or other
[] Connect different modules (classes)
Examples for image classification.
It contains Python training/testing/exporting code, and import C++ code for runtime classification.
NOTE: It requires the CPP version of pytorch in the computer.
I.e. D:\workspace\programs\ThirdPartyLib\libtorch-win-shared-with-deps-1.7.0+cu101
#Additional Link
https://pytorch.org/tutorials/beginner/former_torchies/nnft_tutorial.html
https://pytorch.org/tutorials/beginner/blitz/neural_networks_tutorial.html
https://pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html
https://github.com/utkuozbulak/pytorch-cnn-visualizations
https://github.com/tomjerrygithub/Pytorch_VehicleColorRecognition/blob/master/model.py
https://www.guru99.com/pytorch-tutorial.html
https://datascience.stackexchange.com/questions/67549/validation-loss-is-not-decreasing-regression-model
https://discuss.pytorch.org/t/linear-layer-input-neurons-number-calculation-after-conv2d/28659/5
https://discuss.pytorch.org/t/beginner-should-relu-sigmoid-be-called-in-the-init-method/18689/6
https://stackoverflow.com/questions/57802632/loss-is-not-converging-in-pytorch-but-does-in-tensorflow
https://discuss.pytorch.org/t/sloved-why-my-loss-not-decreasing/15924/6
https://discuss.pytorch.org/t/cnn-does-not-predict-properly-does-not-converge-as-expected/43567
https://discuss.pytorch.org/t/how-to-visualize-the-actual-convolution-filters-in-cnn/13850/11
https://discuss.pytorch.org/t/non-linear-regression-methods/62060/2
https://www.programmersought.com/article/16614780143/
"""
def hingeLoss(outputVal,dataOutput,model):
#print('outputVal:{} dataOutput:{}'.format(outputVal, dataOutput))
loss1=torch.sum(torch.clamp(1 - torch.matmul(outputVal.t(),dataOutput),min=0))
loss2=torch.sum(model.fc2.weight ** 2) # l2 penalty
totalLoss=loss1+loss2
#print('loss1:{} loss2:{}'.format(loss1, loss2))
return(totalLoss)
"""
https://medium.com/udacity-pytorch-challengers/a-brief-overview-of-loss-functions-in-pytorch-c0ddb78068f7
http://alexlenail.me/NN-SVG/LeNet.html
https://medium.com/@RaghavPrabhu/cnn-architectures-lenet-alexnet-vgg-googlenet-and-resnet-7c81c017b848#:~:text=VGG%2D16%20is%20a%20simpler,2%20with%20stride%20of%202.&text=The%20winner%20of%20ILSVRC%202014,also%20known%20as%20Inception%20Module.
https://www.pyimagesearch.com/2017/03/20/imagenet-vggnet-resnet-inception-xception-keras/#:~:text=The%20VGG%20network%20architecture%20was,each%20other%20in%20increasing%20depth.
https://pytorch.org/hub/pytorch_vision_vgg/
https://www.kaggle.com/carloalbertobarbano/vgg16-transfer-learning-pytorch
https://stackoverflow.com/questions/53114882/pytorch-modifying-vgg16-architecture
https://stackoverflow.com/questions/52963909/pytorch-getting-the-correct-dimensions-for-final-layer/52981440#52981440
https://pytorch.org/docs/stable/generated/torch.nn.AdaptiveAvgPool2d.html#torch.nn.AdaptiveAvgPool2d
https://discuss.pytorch.org/t/vgg-16-architecture/27024/4
https://github.com/pytorch/vision/blob/master/torchvision/models/vgg.py
https://pytorch.org/tutorials/beginner/pytorch_with_examples.html
https://discuss.pytorch.org/t/combine-two-model-on-pytorch/47858/6