Deep Learning modeling Repo:

This repository provides a guide for deep learning modeling of different computer vision projects

Overview of deep learning modeling for Mnist digits classification project
-- 70000 digit images from Mnist dataset have been used for multi class classifications
-- Dense neural network has been tested for image classification
-- Convolutional neural network (CNN) used for modeling -- Model Quntization: Quantize model weight using tf.Lite or quantize each model layer using tensorflow_model_optimization, compare the model size

Overview of Transfer learning for flower image classification project
-- 3500 flower photos have been used for multi class classification of flowers between 5 groups:
("roses", "daisy", "dandelion", "sunflowers", "tulips") each containes about 700 images
-- convolutional neural network (CNN) and data agumentaion has been tasted for classification of flowers
-- Transfer Learning with pre-trained model of fishes download from tensorflow.hub used for flower image classification
-- Transfer learning with Inception model used for flower image classification

Overview of Transfer learning for image classification project
-- The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class has been used for image classification project
-- Convolutional neural network (CNN) and data agumentaion has been tested for image classification
-- Transfer learning with ResNet50 model has been used for image classification
-- Transfer learning with MobileNetV2 model has been used for image classification
-- compare accuracy and performance of ResNet50 and MobileNetV2 in image classification

Overview of Transfer learning for tomato leaf disease calssification project
Deep learning modeling of tomato leaf disease detection
Data were taken from : https://www.kaggle.com/datasets/kaustubhb999/tomatoleaf
-- Convolutional neural network (CNN) and data agumentaion has been tested for tomato disease detection
-- Transfer learning with MobileNetV2 model has been used for tomato leaf disease calssification
-- compare accuracy and performance of Transfer learning and pure CNN model in classification

Overview of Vision Transformer (ViT) modeling of image classification project
-- The CIFAR-10 dataset consists of 60000 32x32 colour images in 10 classes, with 6000 images per class has been used for image classification project
--Vision Transformer (ViT) has been applied for CIFAR image classification. Model consists of the following key components:
-- Patch Embedding: Converts an image into a sequence of flattened patches.
-- Position Embedding: Adds positional information to the patches.
-- Transformer Encoder: Applies multiple layers of self-attention and feed-forward neural networks.
-- Classification Head: Maps the final representation to the class labels.
-- compare the accuracy and performance of ViT with pure CNN and other transfer learning in TransferLearning-CIFAR-Project.ipynb

Overview of object detection and localization project with different models
-- Object Detection model on images
-- Tensor Flow object detection models that have been trained on the COCO 2017 dataset, from TensorFlow Hub
-- object detection with YOLO
--compare two models Mask R-CNN Inception and YOLO in object detection

Overview of deep learning models
-- ResNets50 consists of 50 layers of identity_block and Convolutional_block with skip connections -- UNet model (U-shape) consists of an encoder (downsampler) and decoder (upsampler) with a bottleneck in between. It uses the skip connections that concatenate encoder block outputs to each stage of the decoder.