Skip to content
View sghawana's full-sized avatar

Block or report sghawana

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Please don't include any personal information such as legal names or email addresses. Maximum 100 characters, markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
sghawana/README.md

5 Horses


Hi, I'm Sameer Ghawana




πŸ–₯️ Projects

1. Computer Vision and Image Processing

Title Description Tags
Devanagari Character Generator Generate alphabets and numbers from devnagri script using GAN and its different variations GAN, DCGAN, Generative Models
Segmentation of Cricket Images Image segmentation of cricket images, players, equipment, and field elements using a pretrained model. ResNet50, Semantic Segmentation
Object Detection using YOLOv7 I performed the task of object detection on a custom dataset using YOLO v7 and compared the model performance with change in hyper-parameters. YOLO, Object Detection
Fast Convolution Methods Implementation of Depthwise Convolution to speed up CNNs CNN, Convolution
Custom JPEG Implementation I implemented the JPEG image compression algorithm from scratch and compared the compression results with standard library function of OpenCV JPEG, Image Compression

2. Natural Language Processing and LLMs

Title Description Tags
Transliteration of Names I performed the task of sequence-to-sequence mapping using 3 generation of Encoder-Decoder architectures namely RNN, LSTM and Transformers Seq2seq
LSTM
RNN
Custom GPT-2 Tokenizer I built a gpt-2 style tokenizer that can be trained on any .txt data to generate tokens Tokenization
BPE

3. Miscellaneous

Title Description Tags
Classical ML Algorithms Implementation of standard ml algorithms for regression, classification and clustering Regression, Clustering, Classification
NN using Numpy I made a Multi Layer Perceptron(fwd & bkwd pass) entirely in numpy using standard loss and activations MLP, Backpropogation
community-detection Implemented couple of classical graph algorithms for purpose of community detections Graphs, Louvain, Girvan Newman


Paper Implementations:

1. πŸ”  Natural Language Processing and LLMs

Title Repo Description Tags
Efficient Estimation of Word Representations in Vector Space (2013) Link Introduced word2vec (CBOW and Skip-gram) to learn word embeddings that capture semantic relationships of words NLP
Word Embeddings
Word2vec
Neural Machine Translation of Rare Words with Subword Units (2015) Link Introduced BPE Algorithm for sub-word tokenization NLP
Tokenization
Neural Machine Translation by Jointly Learning to Align and Translate (2016) Link Introduced attention for first time in RNN seq2seq architecture NLP
Attention
RNN seq2seq
Attention is all you need (2017) Link First Paper that proposed Transformer architecture to perform sequence to sequence mapping NLP
Transformers
seq-2-seq

2. πŸŒ„ Image Processing, Computer Vision & Generative Models

Title Repo Description Tags
Distinctive Image Features from Scale-Invariant Keypoints Link The original SIFT paper. Describes method for extraction of image features(deterministically), which are invariant to changes in scale, rotation, and illumination Img Processing
Feature Extraction
Noise removal via Bayesian wavelet coring &
Adapting to unknown smoothness via wavelet shrinkage
Link Some Classical Techniques to perform Image denoising using Laplacian and Shrinkage extimates Img Processing
Denoising
Normalized Cuts and Image Segmentation Link Proposes an approach to image segmentation by framing it as a graph partitioning problem and introduces an algorithm to achieve that Img Processing
Segmentation
Fully Convolutional Networks for Semantic Segmentation Link Proposes FCN architecture for the purpose of Image segmentation CV
CNN
Segmentation
Gradient-based learning applied to document recognition Link Proposes Le-Net architecture for the purpose of Document character recogonition CV
CNN
Object detection

Pinned Loading

  1. Devanagri-character-generator Devanagri-character-generator Public

    Generate alphabets and numbers from devnagri script using GAN and its different variations

    Python

  2. Image-segmentation-by-pretrained-ResNet50 Image-segmentation-by-pretrained-ResNet50 Public

    I did semantic segmentation on a custom dataset by the pretrained resNet50 model in PyTorch. I also did a comparative analysis of model(resNet50) configuration with and without skip connections

    Jupyter Notebook

  3. transformer-from-scratch transformer-from-scratch Public

    I built the original transformer encoder and decoder in PyTorch as outlined in the paper "Attention is all you need""

    Python

  4. neural-network-in-numpy neural-network-in-numpy Public

    I made a Multi Layer Perceptron(Neural Network) in numpy

    Python

  5. custom-gpt2-tokenizer custom-gpt2-tokenizer Public

    I built a gpt-2 style tokenizer that can be trained on any .txt data to generate tokens

    Python 1

  6. transliteration-of-Names transliteration-of-Names Public

    I performed the task of sequence-to-sequence mapping using 3 generation of Encoder-Decoder architectures namely RNN, LSTM and Transformers

    Python 1