LIG at TRECVid 2015: Semantic Indexing

LIG participated to the semantic indexing main task. LIG also participated to the organization of this task. This paper describes these participations which are quite similar to our previous year's participations (within the Quaero consortium). Our approach uses a six-stages processing pipelines for computing scores for the likelihood of a video shot to contain a target concept. These scores are then used for producing a ranked list of images or shots that are the most likely to contain the target concept. The pipeline is composed of the following steps: descriptor extraction, descriptor optimization, classification, fusion of descriptor variants, higher-level fusion, and re-ranking. We used a number of different descriptors and a hierarchical fusion strategy. We also used conceptual feedback by adding a vector of classification score to the pool of descriptors. The main innovation this year consisted in the inclusion of semantic descriptors computed using a deep learning method. We also used multiple frames for some features and this did lead to a significant improvement. The best LIG run has a Mean Inferred Average Precision of 0.2935, which ranked it 5$^{th}$ out of 15 participants.