Among the Top 9 at the FashionGen Challenge

Among the Top 9 at the FashionGen Challenge

At the FashionGen Challenge, NUS Electrical Engineering PhD student Kenan Emir Ak’s algorithm, which was based on deep-learning methods, ranked second overall. The FashionGen Challenge was held as part of the Computer Vision for Fashion, Art and Design Workshop at the European Conference on Computer Vision (ECCV 2018) in Munich, Germany (8-14 September 2018).

Top 9 submissions were shortlisted from the many worldwide entries received using algorithm generated scores.  These were then ranked based on user evaluations that checked for visual quality, and the correlation between the text and generated images. Jointly organised by Ssense & Element AI, the  “FashionGen Challenge” is the first of its kind for ‘text-to-image synthesis’ tasks and the winners were determined after which user evaluations was checked for the correlation between the text and generated images.

Kenan is supervised by Prof Ashraf A. Kassim from NUS Electrical & Computer Engineering, Dr Lim Joo Hwee (I2R-A*STAR) and Dr Tham Jo Yew (ESP xMedia Pte Ltd).

An attention-driven generative network for generating fashion images conditioned on input text descriptions developed by a team comprising Mr. Kenan Emir Ak (NUS-ECE & A*STAR scholar), Dr. Ashraf A. Kassim (NUS-ECE), Dr. Lim Joo Hwee (I2R-A*STAR), Dr. Tham Jo Yew (ESP xMedia Pte. Ltd.) was ranked second overall at the recent “FashionGen challenge” organized as part of the Computer Vision for Fashion, Art and Design Workshop at the European Conference on Computer Vision 2018 (ECCV2018) in Munich, Germany (8 to 14 September 2018). The “FashionGen challenge” jointly organized by Ssense & Element AI, is the first of its kind for ‘text-to-image synthesis’ tasks and the winners were determined after user evaluations which checked for the correlation between the text and generated images. Top nine submissions were shortlisted from the many worldwide submissions using algorithm generated scores and these were then ranked based on user evaluations that checked for visual quality, and the correlation between the text and generated images. The figure below shows fashion images generated by our algorithm for various text descriptions.
November 5, 2018

Among the Top 9 at the FashionGen Challenge