Purpose: Use Residual block and create a CNN model for training on CiFAR 10 dataset.

Based on CiFAR 10 dataset

Basic structure for the model:-

• PrepLayer - Conv 3x3 s1, p1) >> BN >> RELU [64k]
• Layer1 -
  X = Conv 3x3 (s1, p1) >> MaxPool2D >> BN >> RELU [128k]
  R1 = ResBlock( (Conv-BN-ReLU-Conv-BN-ReLU))(X) [128k]
  
• Add(X, R1)
• Layer 2 - Conv 3x3 [256k]
  MaxPooling2D
  BN
  ReLU
  
• Layer 3 - X = Conv 3x3 (s1, p1) >> MaxPool2D >> BN >> RELU [512k]
  R2 = ResBlock( (Conv-BN-ReLU-Conv-BN-ReLU))(X) [512k]
  Add(X, R2)
  MaxPooling with Kernel Size 4
  
• FC Layer
• SoftMax Uses One Cycle Policy such that:
  Total Epochs = 24
  Max at Epoch = 5
  LRMIN = FIND
  LRMAX = FIND
  NO Annihilation
  Uses this transform -RandomCrop 32, 32 (after padding of 4) >> FlipLR >> Followed by CutOut(8, 8)
  Batch size = 512
  Use ADAM, and CrossEntropyLoss
  

Project Setup:

Clone the project as shown below:-

$ git clone [email protected]:pankaja0285/era_v1_session10_pankaja.git
$ cd era_v1_session10_pankaja

About the file structure
|__config __config.yaml
|__data
|__data_analysis
|__data_loader
__load_data.py
__albumentation.py
|__models
__model.py
|__utils
__dataset.py
__engine.py
__helper.py
__plot_metrics.py
__test.py
__train.py
|__CiFAR_S10.ipynb
|__README.md

NOTE: List of libraries required: torch and torchsummary, tqdm for progress bar, which are installed using requirements.txt

One of 2 ways to run any of the notebooks, for instance CiFAR_S10.ipynb notebook:

1. Using Anaconda prompt - Run as an administrator start jupyter notebook from the folder era_v1_session10_pankaja and run it off of your localhost
   NOTE: Without Admin privileges, the installs will not be correct and further import libraries will fail.
   

jupyter notebook

2. Upload the notebook folder era_v1_session10_pankaja to google colab at colab.google.com and run it on colab
   

Implementing Residual Block:

File used: models/model.py, model with Net Class and CiFAR_S10.ipynb

Target: - create a model with Residual block

Results:

• Total parameters: 6,573,120
• Train accuracy of 88.14 and test accuracy of 89.38

Analysis:

• To see how the accuracy is, using residual blocks.

Contributing:

For any questions, bug(even typos) and/or features requests do not hesitate to contact me or open an issue!