We are experiencing rapid advancements in AI, especially in the areas of natural language processing and computer vision. These AI breakthroughs have had positive effects on various industries, but the healthcare sector stands to benefit the most.
We have identified several challenges that must be tackled to drive real change in the medical field, particularly concerning medical imaging. The current technology is characterized by a lack of multimodal models, single-task learning, poor explainability, low generalization, and zero dynamism. All these features encourage AI to be viewed as a supplementary tool rather than a critical element in improving patient experiences. Addressing the discussed problem is mandatory to facilitate the adoption of IA technology in health institutions, benefiting the patient's journey and the healthcare teams.
Read MoreSketch-based understanding is a key process in visual perception. Sketching is also a natural and primitive means of communication; it is how our ancestors transmitted ideas, stories, and activities.
Given the critical role sketching plays in visual perception, it is natural to incorporate this process into the field of computer vision. Indeed, we have seen an increasing interest in this topic by the computer vision community, addressing a diversity of sketch-based tasks like sketch classification, sketch representation learning, sketch-guided object localization, sketch-based image retrieval, sketch-to-photo translation, among others. All the sketch-based understanding tasks have gained interest with time, which can be attributed to the development of deep neural networks capable of extracting semantic representations from images, and the ever growing access to smartphones that can be used as a drawing tablet.
Read MoreOne-shot detection involves the capability of a detection model to perform well in front of objects from unseen classes. In this case, the model just knows an example of the item to localize. We are developing efficient models that go beyond a simple image. For example, we combine our experience in the context of sketch-based representations to allow models to localize objects into a catalog of images using a sketch as a query.
Read MoreA generative model is a machine learning model, generally based on deep learning, that aims to learn the distribution of specific data (e.g. images or texts), so that it acquires the ability to generate new data according to the learned distribution. In simple words, it is a model that can generate new images or texts in a specific domain. We can understand generative models as artificial models focused on the synthesis of content. These models have evolved rapidly through variational models (VAE), adversarial networks (GAN) and currently diffusion-based models (DDPM). Our team focused on conditional generative models to produce realistic images guided by an input sketch.
Read MoreVideos are a rich source of multimodal information; they include images, audio and temporal behavior. These features make video analysis a very attractive area as well as a challenge, especially for building an efficient model dealing with such a vast amount of data. Our team is working on models to describe video content in the wild, including news, sports, and streaming. In addition, we also exploit sketches for querying into video catalogs.
Read More