The 1st CV4Africa Workshop will be held this year co-located with Deep learning Indaba 2023. The workshop will feature invited talks from prominent researchers and practitioners, a challenge and hand-on tutorials. We invite all members of the Computer Vision community to attend the workshop. Computer vision is used in various applications that impact African communities such as, precision agriculture, satellite imagery understanding, and medical image processing. It is concerned with the mathematical techniques that include both classical and machine learning based methods towards achieving the goal of scene and video understanding, and recovering the 3D shape and appearance of objects in images. Different sub-tasks in Computer Vision include optical flow, motion detection, tracking, segmentation/grouping, and 3D reconstruction among others. Although, it is widely used to serve our communities, there exists a current gap in the community based research that lacks focus on Computer Vision in Africa. We are mainly inspired by other grassroots initiatives in the African community for both natural language processing (Masakhane) and machine learning for health (Sisonke Biotik). We aim to launch a community which we call Ro’ya-CV4Africa that focuses on Computer Vision research for Africans and by Africans. We believe that this bottom-up community based approach is better able to bring researchers from varying parts of the society and is inclusive by design. Participatory research, unlike conventional research, defines the research process itself within a collaborative and accessible framework to all members of the community.

Spatial Apartheid: Constructing a Visual Dataset to Study the Effects of Spatial Apartheid in SA

Abstract: The lack of ground truth data is usually a barrier when researchers are trying to train machine learning models for real-world applications. Spatial datasets in particular are both complex and expensive to create. In this talk I will be discussing how we constructed a visual dataset that allows us to classify neighbourhoods using machine learning and also allows us to study the effects of spatial segregation in South Africa. While datasets 7 times smaller than ours have cost over $1M to annotate, our dataset was created with highly constrained resources.

Bio: Raesetje is an AI Research Fellow at the Distributed AI Research(DAIR) institute. Her research focuses on creating ground truth datasets and using machine learning and other computational social science techniques to explore research questions with a societal impact. Her current research uses satellite imagery to study the legacy of spatial apartheid in South Africa, post-Apartheid. Her previous work involved partnering with various stakeholders and using machine learning techniques to study poverty and traffic safety in the urban parts of Nigeria and Jakarta, respectively.

Combining Deep Learning and Classical Approaches to Predict Tropical Storms Speeds using Satellite Data

Abstract: Tropical cyclones unfortunately can cause thousands of deaths and billions of dollars of damage, which makes their real time monitoring important to better assess their danger and help save people's lives. In this talk, Karim Amer will discuss his 2nd solution to the "Wind-dependent Variables: Predict Wind Speeds of Tropical Storms" challenge organized by Radiant Earth Foundation and NASA.

Bio: Karim is Co-Founder & CTO, VAIS, a deeptech startup that brings the power of Geo-spatial Intelligence and AI to the hands of African and Egyptian farmers. He has been applying AI for the last six years to challenging interdisciplinary problems in several areas including Agriculture, Earth Observation, Medical Imaging, Material Science, Bioinformatics, and Geophysics. He previously worked at Siemens Healthineers Technology Center, NJ, USA on the development of cutting-edge AI models that can be used in multiple clinical applications which turned into 3 patents (pending approval). Karim won several international AI competitions and has become the first Kaggle Master of Data Science Competitions from Egypt, among a few in the region and top 1% worldwide.

Computer Vision for Remote Sensing: Past, Present and Future

Abstract: Remote sensing (RS) and computer vision (CV) are two rapidly developing fields with the potential to revolutionize our understanding of the world. RS provides information about objects or phenomena without direct contact, while CV extracts useful data from digital images or videos. The combination of CV and RS has led to significant advances in a variety of applications, including earth observation, natural disaster detection, military surveillance, urban planning, crop monitoring, and environmental assessment. In the past, the availability of data and computational resources limited the use of CV methods for RS. However, recent advances in Deep Learning have made it possible to develop more powerful and accurate CV algorithms for RS. The role of CV in RS is expected to continue to grow in the future. The availability of large datasets, pretrained models, and the development of new sensors will enable the development of even more advanced CV algorithms for RS. This talk will provide an overview of the past, present, and future of CV for RS. It will discuss the challenges and opportunities of this field, and highlight some of the recent advances in deep learning for RS, including some work from the Google Accra research lab.

Bio: Abigail Annkah is a Google AI Research Software Engineer in Ghana. She is passionate about using computer vision and optimization to address global challenges, such as bridging information gaps in developing regions, improving health, education, and agriculture. She is also actively involved in mentorship programs and is always looking for opportunities to collaborate with others who share her vision. Her undergraduate studies at the University of Ghana, Legon, in Accra provided her with a foundation in Mathematics and Statistics. She was awarded a full scholarship for a Master of Science in Machine Intelligence from the African Institute of Mathematical Sciences (AIMS) in Kigali, Rwanda. Prior to residency at Google, she worked on several projects, including the Application of Image Analysis to Advanced Complex Microbial Images, Spatio-temporal modeling and Malaria prediction in Sub-Saharan Africa, and Image Denoising using the Diffusion Equations with explicit numerical methods. She worked on mapping built-up environments in Africa using satellite imagery during her residency at Google, and is a contributor to Google's Open Buildings Dataset. In addition to her technical skills, Abigail is a strong advocate for diversity and inclusion in STEM fields. She is an ambassador for Mind the Gap , a Google initiative to encourage female students to pursue careers in STEM, and was mentor for the African Institute of Mathematical Sciences (AIMS) Ghana's Girls in Mathematical Sciences Program (GMSP) in the past. She was a member of a Rwandan team that engaged high school teachers and teacher in-training for problem solving for Mathematics students, and also participated in a pilot version of the Rwandan Mathematical Olympiad.

Panoptic Segmentation with Transformers Tutorial

Domain Generalization Tutorial