1: \begin{abstract}

2: In this paper we address the problem of learning robust cross-domain representations for sketch-based image retrieval 

3: (SBIR). While most SBIR approaches focus on extracting low- and mid-level descriptors for direct feature matching, 

4: recent works have shown the benefit of learning coupled feature representations to describe data from two related 

5: sources. However, cross-domain representation learning methods are typically cast into non-convex minimization 

6: problems that are difficult to optimize, leading to unsatisfactory performance. Inspired by self-paced learning,

7: a learning methodology designed to overcome convergence issues related to local optima by exploiting the samples in a 

8: meaningful order (\ie~easy to hard), we introduce the cross-paced partial curriculum learning (CPPCL) framework. 

9: Compared with existing self-paced learning methods which only consider a single modality and cannot deal with prior 

10: knowledge, CPPCL is specifically designed to assess the learning pace by jointly handling data from dual sources and 

11: modality-specific prior information provided in the form of partial curricula.

12: % CPRL in which a cross-modal 

13: % self-paced partial curriculum learning strategy is designed and

14: Additionally, thanks to the learned dictionaries, we demonstrate that the proposed CPPCL embeds robust coupled representations 

15: for SBIR.

16: %Moreover, with only need of partial-order curricula in learning, CPRL is thus beneficial for large-scale SBIR tasks. 

17: Our approach is extensively evaluated on four publicly available datasets (\ie~CUFS, Flickr15K, QueenMary SBIR and TU-Berlin Extension datasets), showing superior performance over competing SBIR methods.

18: %through two SBIR sub-tasks, namely sketch to image retrieval and sketch-to-face recognition.

19: \end{abstract}

20: