Abstract: When building Computer Vision solutions, emphasis is usually on the modelling side and on leveraging the latest algorithm.
While the model is important, in our experience we found that the key component to deliver a successful solution is to build and maintain a suitable dataset. In the talk, we will distil lessons learned from delivering real-life Computer Vision projects for big organizations that point at this.
In particular, we will discuss:
- disentangling the business goal from the possible technical how-to
- expressing assumptions and controlling for corner-cases through building a suitable dataset
- updating the dataset to reflect new lessons and evolving data
Bio: Andrea has 11y experience in predictive analytics and Machine Learning, having worked and led projects across industries for companies like Shell, Aon, Unilever, Barclays, Mizuho, Network Rails. During his career, he has worked on a number of applications, including financial markets predictions, recommender systems for consumer goods, Computer Vision detection models to prevent theft and digitalize documents, NLP models to automate document parsing and HR predictive analytics. In the last few years, he has been building Remo.ai, a platform to understand, prepare and manage Computer Vision datasets.