my deep learning experiments
SAM(Segment Anything Model) Segment images without training seperatly. Use box/points as guidence
catvsDog.ipynb applies fine tuning and transfer learning on vgg network for cat and dog classification
catvsDog_resnet.ipynb applies finetuned Resnet Model for cat and dog prediction, Which achives accuracy of more than 99% accuracy
flowers_classification_98.4.ipynb applies finetuned Xception network for classifying flowers and achieved more than 98% accuracy
traffic_sign_lenet.ipynb classifies traffic signs(german traffic sign dataset) using a convolutional neural network trained from scratch. Data augmentation helps the classifier to improve the test set accuracy(more than 98%).
Autoencoder.ipynb implements normal autoencoder network(a network that can reconstruct its input) and denoising autoencoder(it reconstructs real input from noisy input data)
VAEcolab.ipynb implements Variational autoencoder network, applied on MNIST dataset
GANcolab.ipynb implements Generative Adverserial network, that reconstructs its input. MNIST dataset is used as input.
GradCamALL.ipynb implements gradcam, which is used to visualize the important parts of image that influenced CNN prediction.This implementaion works with most of the pretrained CNNs available.