Backfilling is the process of re-extracting all gallery embeddings from upgraded models in image retrieval systems. It inevitably spends a prohibitively large amount of computational cost and even entails the downtime of the service. Although backward-compatible learning sidesteps this challenge by tackling query-side representations, this leads to suboptimal solutions in principle because gallery embeddings cannot benefit from model upgrades. We address this dilemma by introducing an online backfilling algorithm, which enables us to achieve a progressive performance improvement during the backfilling process without sacrificing the full performance of the new model after the completion of backfilling. To this end, we first show that a simple distance rank merge is a reasonable option for online backfilling. Then, we incorporate a reverse transformation module for more effective and efficient merging, which is further enhanced by adopting metric-compatible contrastive learning. These two components help to make the distances of old and new models compatible, resulting in desirable merge results during backfilling with no extra computational over-head. Extensive experiments show the benefit of our frame-work on four standard benchmarks in various settings.

We present a Reinforcement Learning (RL) solution to the view planning problem (VPP), which generates a sequence of view points that are capable of sensing all accessible area of a given object represented as a 3D model. In doing so, the goal is to minimize the number of view points, making the VPP a class of set covering optimization problem (SCOP). The SCOP is NP-hard, and the inapproximability results tell us that the greedy algorithm provides the best approximation that runs in polynomial time. In order to find a solution that is better than the greedy algorithm, (i) we introduce a novel score function by exploiting the geometry of the 3D model, (ii) we device an intuitive approach to VPP using this score function, and (iii) we cast VPP as a Markovian Decision Process (MDP), and solve the MDP in RL framework using well-known RL algorithms. In particular, we use SARSA, Watkins-Q and TD with function approximation to solve the MDP. We compare the results of our method with the baseline greedy algorithm in an extensive set of test objects, and show that we can outperform the baseline in almost all cases.

Segmentation of brain structures from MR images is crucial in understanding the disease progress, diagnosis, and treatment monitoring. Atlases, showing the expected locations of the structures, are commonly used to start and guide the segmentation process. In many cases, the quality of the atlas may have a significant effect in the final result. In the literature, commonly used atlases may be obtained from one subject's data, only from the healthy, or depict only certain structures that limit their accuracy. Anatomical variations, pathologies, imaging artifacts all could aggravate the problems related to application of atlases. In this paper, we propose to use multiple atlases that are sufficiently different from each other as much as possible to handle such problems. To this effect, we have built a library of atlases and computed their similarity values to each other. Our study showed that the existing atlases have varying levels of similarity for different structures.

This paper presents a new active contour-based, statistical method for simultaneous volumetric segmentation of multiple subcortical structures in the brain. In biological tissues, such as the human brain, neighboring structures exhibit co-dependencies which can aid in segmentation, if properly analyzed and modeled. Motivated by this observation, we formulate the segmentation problem as a maximum a posteriori estimation problem, in which we incorporate statistical prior models on the shapes and intershape (relative) poses of the structures of interest. This provides a principled mechanism to bring high level information about the shapes and the relationships of anatomical structures into the segmentation problem. For learning the prior densities we use a nonparametric multivariate kernel density estimation framework. We combine these priors with data in a variational framework and develop an active contour-based iterative segmentation algorithm. We test our method on the problem of volumetric segmentation of basal ganglia structures in magnetic resonance images. We present a set of 2-D and 3-D experiments as well as a quantitative performance analysis. In addition, we perform a comparison to several existent segmentation methods and demonstrate the improvements provided by our approach in terms of segmentation accuracy.