Tasks and Deliverables
Task 1 Unsupervised Bio-Inspired Image Representations
- D1.1 (T0 + 18): State-of-the-art report on saliency models and biologically-inspired networks. [PDF]
- D1.2 (T0 + 18): Technical report + code on new saliency inspired image descriptors. [PDF]
- D1.3 (T0 + 18): Technical report + code on the hybrid biologically-inspired method. [PDF]
Task 2 Interactive Eye-driven Learning System
- D2.1 (T0 + 18): State-of-the-art report on representation learning and instance-basedmethods including eye-tracker. [PDF]
- D2.2 (T0 + 18): State-of-the-art report on gaze features for eye-tracker based CBIR systems with active learning strategies. [PDF]
- D2.3 (T0+30): Report on using gaze features for weakly supervised learning. [PDF]
Task 3 Case Study: Web Filtering For Food Recipe Annotation
- D3.1 (T0 + 18): Report on food dataset harvesting, cleaning and organization. [PDF]
- D3.2 (T0+30): Report on deep learning for food recipes classification. [PDF]
Upcoming reports:
- D2.5 - Interactive strategy (Thome, All) [T0+42]
Final report + code of new interactive strategy with gaze feature criteria (active learning strategies for SVM binary classifiers / exploit saccadic movements as a binary decision to refine the classifier / analysis of the gaze direction, of pupil parameters, and of eye trajectories can be used as relevance feedback signals instead of a mouse click / short relevance feedback loop / inspiration from Oyekoya’s active learning strategy)
- D3.3 - Scalable CBIR (Precioso, I3S) [T0+42]
Report on scalable interactive hybrid SVM-IEC CBIR system
- D3.4 - Final Interactive model (Thome, All) [T0+48]