Data processing and analysis

The revolution linked to artificial intelligence in image, video and sound processing has greatly modified the techniques developed within our team. The work of the members of the TAD team converges on deep learning. This work has given rise to scientific contributions in various fields: data augmentation, processing of massive data, learning in a context where little data is labelled, or collaborative learning.

Our research work focuses on the following areas:

Data fusion. Our work focuses on the fusion of various data: video, sound, LIDAR, image, MRI, text, multimedia, etc. Our advanced deep learning methods have enabled the reconstruction of ancient artefacts (papyri, vases, skulls). The team has proposed original work on the accurate reconstruction of urban environments from multimodal data combining laser and optical sources. We have also proposed a hybrid time model to merge static and real time data into an interactive multimedia score.

Data processing Several projects include data processing and restoration. For example, our new methods for colourising black and white photos and videos allow for realistic rendering. We have also worked on the completion of missing data in LiDAR acquisitions. 

Knowledge extraction :

We have recognised expertise in the field of segmentation, object and event detection, classification, data mining, etc. We are working on intelligent methods for the extraction of knowledge.

We work on artificial intelligence and signal processing methods. For example, our 3D segmentation methods are applied to medical images. Our 2D segmentation methods are applied to the segmentation of LIDAR data, clothing or faces. We also offer tools for sound localisation and perception-based music analysis. Our work also focuses on the detection of cognitive states such as drowsiness or depression based on voice analysis.

Knowledge production: 

We are developing classification and artificial intelligence algorithms ranging from medical diagnosis to music recommendation.

We contribute to the development of image databases: natural images, medical images, document images, etc.

Finally, we have a strong expertise in massive data analysis. The results of our work in music audio signal processing have led to the development of industrial solutions processing catalogues of tens of millions of titles, used by many international companies in the digital music field.

Link to our research groups :