This project acts as a pipeline to perform certain image transformations as a method to preprocess images, and feeding those transformed images though CNN models. Depending on the input classes, different features will be extracted, and a good predictive model can be determined for the usecase.
Four transforms are used in this project :
Four pre-trained models (with learnable and un-learnable parameters) have been implemented as well :
This project has the following features :
- The input can contain any number of classes. The models will adjust accordingly.
- The model has been deployed on a web-based UI hosted on streamlit for ease of use.
- We get a learning graph at the end as outputs for comparing the learning of the models form these four models.
The input should be a .zip file containing number of folders the same as the number of classes in the dataset. Each of those folders need to contain images of those classes.
The output is in the form of learning graph showing learning curve of the transformed images as the base data.
- Windows 10 64-bit.
- Python 3.10.2
- Nvidia GTX 1050m
- Clone this repository using
git clone https://github.com/accelbia/CV-Project-Wavelet-Transform-Prediction-Model-Analysis. - Navigate to the current cloned repository.
- Install the necessary requirements using
pip install -r requirements.txt. - Run the streamlit application by running
streamlit run Home.py.
-
Upload the dataset by clicking on the 'Browse files'.
-
The classes contained in the zip files are displayed. Choose the model you want to use and add in the relevant layers and information.
-
Choose the optimizer, loss function and metrics, and click start to start the training.
-
The training information is displayed on the screen after the training.
CB.EN.U4CSE19106 - Ayush Barik
CB.EN.U4CSE19106 - T. Ganesh Rohit Sarma
CB.EN.U4CSE19106 - A Shivaani
CB.EN.U4CSE19106 - V Neelesh Gupta