Datasets are complex and hard to visualise – Viz-EDA is an exploratory data analysis tool that helps to see through the data.
Computer vision requires labelled datasets – these are often filled with errors and “anomalies” that can destroy the performance of models trained on them. The image labelling industry is booming, but how can we ensure quality of labelled images without checking each and every one? Here comes Viz-EDA the automatic anomaly and data visualisation tool for industry. The tool exists as a simple Flask app but you can build it further or start from scratch working with a real industry client.
The project will involve understanding the different types of CV datasets, building a modular visualisation tool that runs with both local and cloud-hosted data. Finally, an automated “anomaly” detector using a trained computer vision model to check the data.