Abstract:
The current multi-class food recognition is extremely successful with numerous studies showing 80%-90% of top-1 accuracy is achievable on large image dataset. However, scaling up the recognition from few hundreds to thousands of food categories remain very difficult, due to requirement of tedious efforts in data collection, cleaning and labeling. Furthermore, as the popularity of food categories is often a long-tail pattern, large-scale multi-class training is expected to suffer from data imbalance problem. To bypass these problems, food classification is recently addressed from a different perspective by retrieving the recipes of food for recognition. Specifically, this is a cross-modal retrieval problem where the input is a food picture and the output is a recipe describing food preparation. With recipe retrieval, recognition is not limited to food category, but extraction of rich food attributes such as ingredients and cooking methods becomes possible. More importantly, the cost of training data collection can be significantly cut short, by requiring only the labeling of image-recipe pairs. This talk will share with you the recent progresses on these two methodologies: multi-class multi-label food recognition and recipe retrieval based recognition. The talk will discuss the pros and cons of these methodologies and outline few potential challenges towards scalable food recognition.

Bio:
Chong-Wah Ngo is a professor in the Department of Computer Science at the City University of Hong Kong. He received PhD in Computer Science from Hong Kong University of Science & Technology, and MSc and BSc, both in Computer Engineering, from Nanyang Technological University of Singapore. Before joining City University of Hong Kong, he was a postdoctoral scholar in Beckman Institute at the University of Illinois in Urbana‐Champaign. His main research interests include large‐scale multimedia information retrieval, video computing, multimedia mining and visualization. He was the associate editor of IEEE Trans. on Multimedia and is currently the steering committee member of TRECVid, ICMR (Int. Conf. on Multimedia Retrieval) and ACM Multimedia Asia. He is program co-chair of ACM Multimedia 2019, and general co-chairs of ICIMCS 2018 and PCM 2018. He served as the chairman of ACM (Hong Kong Chapter) during 2008-2009, and was named ACM Distinguished Scientist in 2016 for contributions to video search and semantic understanding.

Abstract:
Recipes often capture ingredients and processes for making specific foods and meals. Yet they are often unable to capture nuances of methods, as well as observations of intermediate stages and expected outcomes. In addition to recipes, culinary knowledge has traditionally been passed to successive generations via a combination of observation, practice, and oral communication. Modern digital communications continue to intermediate several types of interpersonal communications, including communication of traditional cooking knowledge. Unlike previous periods in human history, humanity is at great risk for losing culinary knowledge and cultural heritage at a global scale. Prioritizing the preservation of food knowledge holds implications for cultural heritage and identity, food security, and of course, discovery and deliciousness. Fortunately, the same portable digital access devices like phones that currently intermediate interpersonal communications also provide an opportunity to collect and access verbal and nonverbal communications around food in queryable multimedia formats. Burgeoning collaborations across digital sciences and humanities afford community leveraging of agriculture, food, eating and health ontologies to link objective data with subjective experiences in order to preserve and share the world’s culinary identities and food ways evolved over millennia

Bio:
Chef Robert Danhi’s current mission is to curate, code, archive and share the worlds culinary cultures hence he has created, Flavor360, a software platform with a family of apps for used to blend qualitative and quantitative research methodologies and document the 360° Flavor Experience. The Flavor360 Solutions team partners with organizations to efficiently capture and share the tribal knowledge of their organization during every stage of the innovation, R&D, production and launch process. Robert is also owner of Chef Danhi & Co Consulting, s a full service agency specializing in menu & product R&D, sales & marketing support, educational training programs for restaurants, food manufacturers, marketing associations and academic organizations. Robert is the award winning author of the James Beard Finalist book Southeast Asian Flavors and Easy Thai Cooking, host of the 26-episode docuseries, Taste of Vietnam and a main judge on Top Chef Vietnam.   www.chefdanhi.com

The participants are asked in a practical session to elaborate descriptors for describing food products and discuss the terms in order to find references and

definitions. This is done in small groups so potential cross-cultural influences are highlighted. Findings are presented in the plenum.The objective of this lecture is to

demonstrate the challenges in verbally exchange taste experiences.

4:30 - 4:45 pm.

DECOR2019 Best Paper Award.