NLPR was awarded the 3rd place in 2013 MSR-Bing Image Retrieval Challenge
Mar 14, 2014Author:
PrintText Size A A

The 2013 MSR-Bing Image Retrieval Challenge Workshop (MSR-Bing IRC 2013) was held at Bellevue, WA, USA on Oct 7, 2013. Among the 39 teams, the team of NLPR_MMC, led by Prof. Changsheng Xu from National Lab of Pattern Recognition (NLPR) was awarded the 3rd place. Microsoft Executive Vice President, Harry Shum, presented the award.

NLPR_MMC team has proposed a framework combining search-based and model-based solutions. The framework exploits several machine learning algorithms towards the practical image retrieval applications, to obtain the balance between accuracy and efficiency. Under the precondition to ensure the retrieval efficiency, comparing with the baseline provided by Bing, the proposed framework achieves 8% accuracy improvement. The satisfactory generalization capability is highly recognized by Microsoft research and development sectors.

The 2013 MSR-Bing Image Retrieval Challenge is jointly held by Microsoft Research and Bing, which aims to bring researchers and practitioners to present and discuss large-scale image retrieval algorithms and systems. The challenge provides one million real web images, with attached original user query words and click records. Without using external data resources, each team is required to submit an online system to evaluate the relevance score of query word and images. Comparing with traditional image classification or image annotation problems, image retrieval is most challenging in that both original query words and web images are required to be considered simultaneously. Being the first large-scale image retrieval challenge, MSR-Bing IRC is expected to play a more impactful role on both industry and academia.