In this series of liveProjects, you’ll join up with four different computer vision companies to explore computer vision models powered by the latest deep learning architectures. You’ll utilize the groundbreaking transformer architecture, which forms the driving force behind ChatGPT, to develop a series of increasingly complex models. Starting with a classifier to detect brain tumors, you'll move on to a segmentation algorithm and an object detection application capable of detecting construction vehicles and structural flaws. Finally, you’ll take on the role of an MLOps expert, implementing model deployment and explainability in the systems you’ve developed.
In this liveProject, you’ll take on the role of an engineer at AISoft, where you'll be part of two dynamic teams: MLOps and Core-ML. On the MLOps team, you'll utilize software engineering techniques to ensure your models are not only accurate but also scalable, reliable, and maintainable. You’ve been assigned two important support tasks for the Core-ML team.First, you'll utilize the Gradio Python library to create an interactive web user interface that runs MRI classification and segmentation transformer models in the background, and then you'll build a pipeline that provides interpretability to the decisions of a construction vehicle classification model.
In this liveProject, you'll spearhead the development of AI-aided surveillance software for construction site supervision. You’ll build two computer vision applications capable of detecting construction vehicles and their types across a large worksite and a more powerful model that can detect building defects such as cracks and fissures. Start by working with a pre-trained DETR model, then explore the Roboflow platform to assist you as you create a multi-class object detection dataset from multiple datasets with non-identical classes. With these datasets, you will train different transformer models for object detection to identify construction vehicles and cracks in buildings.
In this liveProject, you'll pioneer the development of cutting-edge MRI segmentation algorithms using transformer architecture for computer vision company VisionSys. Manual segmentation is labor-intensive and expensive, so you’ll be developing a custom model that can do it for you. You'll train and evaluate SegFormer and MaskFormer models to identify brain tumor regions with over 90% accuracy. With Python tools like Hugging Face Transformers and Google Colab's GPU computing resources, you'll create pipelines, preprocess data, and showcase sample predictions and quantitative results.
In this liveProject, you’ll join BrainAI’s MRI data analysis team. BrainAI needs you to develop a state-of-the-art AI module utilizing vision transformer models to detect brain tumors with exceptional accuracy. Armed with Python tools like Hugging Face Transformers, PyTorch, and more, you'll detect the presence or absence of tumors within human-brain MRI datasets. With Google Colab's GPU computing resources, you'll utilize deep learning to try and achieve a 95%+ accuracy rate.
In this series of liveProjects, you’ll use deep learning to build an image classification model that can perform early diagnostics of COVID-19. Your model will examine X-rays and CT scans and seek to classify them as either COVID or non-COVID types. You’ll start by training custom deep learning models from scratch, and then experiment with transfer learning using premade models. Familiarity with both these approaches will ensure you’re able to design a powerful solution for any deep learning problem, model, or dataset. Finally, you’ll test and report on the efficiency of trained models to decide which model delivers the best results.
In this liveProject, you’ll implement model performance metrics that can test the effectiveness of your models. You’ll calculate accuracy, precision, F1 score and recall values from the classification results for an existing model, and then estimate the ROC curve and AUC value. Finally, you’ll create a Gradient Class Activation Map. This map can highlight features and regions in an image that the deep learning model finds important, and manually inspect whether the model is performing in the desired way.
In this liveProject, you’ll take pretrained VGG16 and ResNet models from the Python Keras library and train them further upon your medical image dataset of X-ray and CT scans. This transfer learning is a highly effective technique for quickly generating reliable machine learning models when you only have a small data set. You’ll experiment with the Keras loss functions to determine which are best for COVID image classification, and check your training and prediction times as a critical parameter of real-world applications.
In this liveProject, you’ll build a ResNet deep learning model from scratch to analyze medical imagery. A ResNet is a deep neural network model which uses "Residual blocks" and "skip connections" to reduce the need for very deep networks while still achieving high accuracy. You’ll then train your model on X-ray and CT datasets, and plot validation loss, and accuracies vs. epochs. You’ll build an important familiarity with the functional blocks of a DL model, how data must be formatted, and which layers to use to solve your problems.
In this liveProject, you’ll build a VGG16 deep learning model from scratch to analyze medical imagery. A VGG16 is a deep convolutional network model which has shown to achieve high accuracy in image based pattern recognition tasks. You’ll then train your model on X-ray and CT datasets, and plot validation loss, and accuracies vs. epochs. You’ll build an important familiarity with the functional blocks of a DL model, how data must be formatted, and which layers to use to solve your problems.