EML-NET
In this work, we apply state-of-the-art Convolutional Neural Network(CNN) architectures for saliency prediction. Our results show that better saliency features can be delivered by a deeper CNN model. However, it is very space-consuming to apply a complex model due to the large size of input images. The space complexity becomes even more problematic when we extract features from multiple convolutional layers or different models. In this paper, we propose a modular saliency system which aims at splitting the whole network into small modules. The main difference in our approach s that the encoder and decoder can be separately trained for the scalability. Furthermore, the encoder can contain more than one CNN model to extract features and the models can have different architectures or pre-trained on different datasets. This parallel design allows us to better utilize the computational space in order to apply more powerful encoder. More importantly, our network can be easily expanded almost without extra spaces, other pre-trained CNN models can be combined for a wider range of visual knowledge. We denote our expandable multi-layer network as EML-NET in this paper. Our method is evaluated on three public saliency benchmarks, SALICON, MIT300 and CAT2000. The proposed EML-NET achieves state-of-the-art results on the metric of Normalized Scanpath Saliency using a modified loss function. …

N-Stars Network Evolution Model
A new network evolution model is introduced in this paper. The model is based on co-operations of $N$ units. The units are the nodes of the network and the co-operations are indicated by directed links. At each evolution step $N$ units co-operate which formally means that they form a directed $N$-star subgraph. At each step either a new unit joins to the network and it co-operates with $N-1$ old units or $N$ old units co-operate. During the evolution both preferential attachment and uniform choice are applied. Asymptotic power law distributions are obtained both for the in-degrees and the out-degrees. …

Pairwise Inner Product (PIP)
In this paper, we provide a theoretical understanding of word embedding and its dimensionality. Motivated by the unitary-invariance of word embedding, we propose the Pairwise Inner Product (PIP) loss, a novel metric on the dissimilarity between word embeddings. Using techniques from matrix perturbation theory, we reveal a fundamental bias-variance trade-off in dimensionality selection for word embeddings. This bias-variance trade-off sheds light on many empirical observations which were previously unexplained, for example the existence of an optimal dimensionality. Moreover, new insights and discoveries, like when and how word embeddings are robust to over-fitting, are revealed. By optimizing over the bias-variance trade-off of the PIP loss, we can explicitly answer the open question of dimensionality selection for word embedding. …

Localization Recall Precision (LRP)
Average precision (AP), the area under the recall-precision (RP) curve, is the standard performance measure for object detection. Despite its wide acceptance, it has a number of shortcomings, the most important of which are (i) the inability to distinguish very different RP curves, and (ii) the lack of directly measuring bounding box localization accuracy. In this paper, we propose ‘Localization Recall Precision (LRP) Error’, a new metric which we specifically designed for object detection. LRP Error is composed of three components related to localization, false negative (FN) rate and false positive (FP) rate. Based on LRP, we introduce the ‘Optimal LRP’, the minimum achievable LRP error representing the best achievable configuration of the detector in terms of recall-precision and the tightness of the boxes. In contrast to AP, which considers precisions over the entire recall domain, Optimal LRP determines the ‘best’ confidence score threshold for a class, which balances the trade-off between localization and recall-precision. In our experiments, we show that, for state-of-the-art object (SOTA) detectors, Optimal LRP provides richer and more discriminative information than AP. We also demonstrate that the best confidence score thresholds vary significantly among classes and detectors. Moreover, we present LRP results of a simple online video object detector which uses a SOTA still image object detector and show that the class-specific optimized thresholds increase the accuracy against the common approach of using a general threshold for all classes. We provide the source code that can compute LRP for the PASCAL VOC and MSCOCO datasets in https://…/LRP. Our source code can easily be adapted to other datasets as well. …