Only look once, mining distinctive landmarks from ConvNet for visual place recognition

Abstract

-Recently, image representations derived from Convolutional Neural Networks (CNNs) have been demonstrated to achieve impressive performance on a wide variety of tasks, including place recognition. In this paper, we take a step deeper into the internal structure of CNNs and propose novel CNN-based image features for place recognition by identifying salient regions and creating their regional representations directly from the convolutional layer activations. A range of experiments is conducted on challenging datasets with varied conditions and viewpoints. These reveal superior precision-recall characteristics and robustness against both viewpoint and appearance variations for the proposed approach over the state of the art. By analyzing the feature encoding process of our approach, we provide insights into what makes an image presentation robust against external variations.

Learning Context Flexible Attention Model for Long-Term Visual Place Recognition

“…Similarly, in [26] the attention mechanism was explored to identify salient landmarks for mobile robot localization. Approaches in [8,9] identify salient regions using an external landmark detector. However, their landmark detectors are based on networks that are trained on tasks different in nature from place recognition.…”

“…Similar to [9], we treat activations at a certain convolutional layer as a tensor of size H × W × C, which is considered as a set of C-dimensional local features in H × W spatial location. Formally, the convolutional layer activations X ∈ × × can be expressed as:…”

“…The approaches proposed in [8,9] first identify salient regions and then extract features to match image patches over extreme condition-and viewpoint-variations. However, these methods interpret saliency selection and feature extraction as two separate processes, while our model can optimize both processes simultaneously in an end-to-end fashion.…”

Attention-Aware Age-Agnostic Visual Place Recognition

“…[22] proposes a global image representation by the regional maximum activation of convolutional layers (R-MAC) well-suited for place recognition. [2] proposes novel CNN-based features designed for place recognition by detecting salient regions and extracting regional representations as descriptors. NetVLAD [1] introduces a novel triplet ranking loss together with a VLAD aggregation layer that can learn powerful representations for the VPR task in an end-to end manner.…”

Spatio-Semantic ConvNet-Based Visual Place Recognition

“…Their results on some very challenging datasets were remarkable, significantly outperforming state-of-the-art works based on pre-trained CNNs. Recently, following the success of region-based approaches, Chen et al [23] created a system that used a late convolutional layer as a landmark detector and an earlier one to create local descriptors to match the detected landmarks. Their system showed improved recognition under strong viewpoint and condition variations.…”