r0han99 / deep-learning-anatomy Goto Github PK

A progressive web application which is essentially the dissection of Deep Learning with a diverse spectrum of projects. Post every project's completion, all the meta-data unique to that deep learning model of a particular problem statement is dynamically collected, organized, and byte-encoded into a file to establish the project parameters for future reference and better comprehension of the project. This organized information from the file is then accessed and rendered on the web application. Real-time evaluation of the Model constructed is also made available. This web catalog also has a "Reference Section," (dubbed as Anatomy ) which can be treated as a resource for any novice learner to "Interactively Experience" the concepts like Gradient Descent, Image Decomposition into Features with various levels of Convolutions, Various Metrics & their usage. The Streamlit Library facilitated the front-end aesthetics with complete python codebase.

( Pre-Deployment )

• MNIST
• CIFAR-10

( Post-Deployment )

• Dogs-&-Cats

• MNIST : Anchores the process of establishing the comprehensive report sheet which is to be rendered on the web-page.

• CIFAR-10 : Facilitates the process of Constructing and Testing a robust pipeline script which dynamically creates report-sheets with minimal human intervention.

• Dogs-&-Cats : Validates the report generation process and pipeline robustness post-deployment

Note, These 3 projects were the support structures to test and automate the process to fill the project catalogue dynamically.

Convolutional Neural Networks

Iteration-0

• MNIST
• CIFAR10

Transfer Learning

Iteration-0

• Galaxy Classifier ( InceptionV3 )
• Dogs and Cats Classifier (MobileNetV2)

• [6th Feb, 2021] Deployed
  • A seamless automated deployment on streamlit-share

• [24th Jan, 2021] init

  • Developed a pipeline script which is executed post project development, this gathers all the meta information from the project such as The Neural Network Architecture ( layers, Hidden Units ), Type of Prediction ( Classification, Regression), Type of Network ( Convolutional, Recurrent ) etc. The meta info is organised and a report file is created, which is then processed by the application from the project dir to be rendered on the Web Page.
  • Partially developed Application codebase. Front-End Project Artefacts are laid down

• [26th Jan, 2021] new patches

  • created variable_dict and prompt_dict
  • Added commandline functionalities to the pipeline script to handle var-dicts and prompt-list
  • now the prompts under the project artefacts are predefined and dynamically generated
  • segmented artefacts and plots into radios
  • added download-link to the source-code pdf
  • cs_plots() will render the plots related/included in a particular project

continuation... ( last-date in changelog - [10th March, 2021] )