Mr. Okpor Samuel's repositories
2D-Measurement-using-Opencv
in this series i use omly opencv to measure 2d areas
Attention-Mechanism-Optimized-ResNet50
in this project i use resnet 50 as the base model, i include Feature fusion layers and then add CBAM mechanisms after the fused layers, this showed a 2% improvemet over the feature fusion model and also a 6% improvement over the base resnet50 model
Build-Your-Custom-GPT-with-LangChain
use the langchain Api to build your very own gpt application and render it to any server
Feature-Fusion-ResNet50
In this project i modify the base resnet 50 with Concatenated Feature Fusion layers, this alone increases precision over the base model by 4%
A-googleNet-Inception-V2-classifier
in this project i use the deprecated Inceptionv2 to build a classifier, the classifier uses a categorical entropty to classify only two items. this shows how the categorical entropy can both be used for binary classification and also multi-class classification
A-ResNet18-ResNet34-ResNet50-ResNet101-ResNet152-Classifier
In this project, i break down how you can use all the ResNet Algorithms to perform classification on a custom dataset in the tensorflow frame work. all models are built from scracth and could be used as you like
A-VggNeT-Classifier
In this project i use a VGG 16 model to classify 3 images, following the process it can basically classify any images from 0-1000 classes. my entire code and thought process has been uploaded
AdAS---Vehicle-Vision-Perception
In this series i will attempt to use machine learning to develop a stable and relaible ADAS system from scratch. this will be a long series and not certain how long it will take me. i hope from this repository you get to understand how ADAS systems are being built from scratch
AI-supervises-AI-Reinforcement-learning
in this project i use A RL agent to supervise or judge the accuracy of a FasterRCNN deep learning network. this was originally made for Defect inspection to see if the NG parts were actually Ng hence reducing the rates of false positives.
Base-ResNet101-vs-Modified-ResNet-101-with-Feature-Fusion-
in this project i use the resnet model to perform classification on 5 different items, the models shows promising results
Cancer_Segmentation_API
in this project i use fastapi to build an easy image segmentation instance that can be used on any image. in this case study for cancerous cells
generate-3D-masks-from-2D-images
In this project i will show you a high level format of how you can transform 2d images into 3d, note that this was a test interview project and the exact parameters where not given to me, but i will show you how i managed to scale the interview
Introduction_to_NLP_using_Spacy-library
in this project i use the easiest and most reliable python library for Natural language processing. this is an easy to use implementation that takes you over the basics and an applied example to analyze a financial document
LDA-for-natural-language-processing
in this project i use Latent dirichlet allocation to perform Gibbs sampling of text on a given dataset
LiDar-3D-object-detection-using-Yolov4-Complexer
Lidar sensors are the best way to perform 3d detection from point clouds, in this project i implement the use of Yolov4 complexer to perform 3d detection
Midas-Depth-Estimation
Midas is a Machine Learning library for depth estimation, in this work i show how to use midas library in pytorch frame work to perform 3D depth estimation using your camera
Object-detection-Faster-Rcnn-Pytorch
in this project i build upon the lapses and fixed the errors ecountered in the suture thread detection to make a proper object detection program than can be used on any dataset of your choice.
Poissoin-3D-reconstruction-using-landmarks
in this project i use media pipe landmarks to convert 2d images into 3d
PyTorch-Image-Segmentation-with-UNET
Image segmentaion is a critical domain in computer vision in the development of autonomous perception. in this project, i go about the development of the UNet encoder decoder architecture and how to use it to perform image segmentation.
q-Learning-Reinforcement-Learning
q learning is a reinforcement algorithm that enables a model perform the best course of action, in this project ill show you how to implement in a simple snake game environment
read-a-data-matrix-opencv-python
reading data matrix is quite cumbersome unlike reading qr codes, in this project i show you the best implementation to read and display a data matrix using a Message Box
ResNet152-classifier
in this project i use the resNet152 as a model instance for classification, this comes the end of bulding resNet CNN architectures models based on skip connections
supervised-LDA
this it the python equivaalent of slda for topics selection, mostly hardcoded in python and using similar R language libraries
Suture-thread-Detection-using-Pytorch-
In this project, i perform small object detection using Faster r-cnn
Vehicle-Vision-Odometry
in this project i try to recreate how certain autonomous vehicle use computer vision techniques to map out the vehicles movement, along a spatial axis. this code should not be implemented into an autonomous vehicle as there are more factors needed for full Autonomous driving. but it can be your starting point
NANO-GPT-LLM-on-Custom-Dataset
in the project i develop and improve AndreJ Karpathy source material on building Gpt1, this model is currently under construction
Skin-Sensitivity-Color-detection-and-Severity-Grading
in this project, imagine the red sensitive areas on a human skin, and we want to first identify them using a color space to highlight only the red areas and also grade the severity of the red skin present. 99~ 75 is mild range, 77~ 40 is the moderate range and 39~1 is the Severe range.