Sports-Conditional Exercise Image Binary Classifier

This project is part of internship assessment test by i'm beside you. Dataset and task description was given the company itself.

Summary

Importing libraries, data, and pre-processing
- Importing Libraries
- Mounting Gdrive
- Extracting Zip file
- Exploring un-zipped directories
- Pre-processing images into batches and resizing
Approach 1: Making our own CNN
- Initial Model (Same as tiny VGG, 82.47% Accuracy)
- Model-2 (Increasing trainable parameters of initial, 81.30% Accuracy)
- Training Initial model for more epochs (82.77% Accuracy)
Approach 2.1: Transfer Learning (Feature Extraction)
- Model 1: Resnet_v2_50 (91.72% Accuracy)
- Model 2: Resnet_v2_152 (91.82% Accuracy)
- Model 3: Efficientnet_B0 (92.79% Accuracy)
- Model 4: EfficientNet_v2 (93.77% Accuracy)
- Model 5: EfficientNetB3-sports-0.97 (67.19% Accuracy)
Approach 2.2 Transfer Learning (Fine Tuning)
- EfficientNetB3-sports-0.97
  - Making Layers trainable
  - Normal training data: 94.16%
  - Augmented training data: 95.72%
- EfficientNet_v2
  - Making Layers trainable
  - Normal training data: 95.72%
  - Augmented training data: 96.69%
  - Training for more Epochs and saving checkpoint file: 97.76%
Saving the model