Zoom Out-and-In Network with Map Attention Decision for Region Proposal and Object Detection

Abstract

In this paper, we propose a zoom-out-and-in network for generating object proposals. A key observation is that it is difficult to classify anchors of different sizes with the same set of features. Anchors of different sizes should be placed accordingly based on different depth within a network: smaller boxes on high-resolution layers with a smaller stride while larger boxes on low-resolution counterparts with a larger stride. Inspired by the conv/deconv structure, we fully leverage the low-level local details and high-level regional semantics from two feature map streams, which are complimentary to each other, to identify the objectness in an image. A map attention decision (MAD) unit is further proposed to aggressively search for neuron activations among two streams and attend the most contributive ones on the feature learning of the final loss. The unit serves as a decisionmaker to adaptively activate maps along certain channels with the solely purpose of optimizing the overall training loss. One advantage of MAD is that the learned weights enforced on each feature channel is predicted on-the-fly based on the input context, which is more suitable than the fixed enforcement of a convolutional kernel. Experimental results on three datasets, including PASCAL VOC 2007, ImageNet DET, MS COCO, demonstrate the effectiveness of our proposed algorithm over other state-of-the-arts, in terms of average recall (AR) for region proposal and average precision (AP) for object detection.

Imbalance Problems in Object Detection: A Review

“…Different from Kong et al [74] and Li et al [75], Multi-Level FPN [76] stacks one highest and one lower level feature layers and recursively outputs a set of pyramidal features, which are all finally combined into a single feature pyramid in a scale-wise manner (see Figure 10(g)). Feature fusion module (FFM) v1 equals the dimensions of the input feature maps by a sequence of 1 × 1 convolution and upsampling operations.…”

Deep Learning for Generic Object Detection: A Survey

“…This approach has been found to be effective for segmentation [177,241] and human pose estimation [194], has been widely exploited by both one-stage and two-stage detectors to alleviate problems of scale variation across object instances. Representative methods include SharpMask [214], Deconvolutional Single Shot Detector (DSSD) [77], Feature Pyramid Network (FPN) [167], Top Down Modulation (TDM) [247], Reverse connection with Objectness prior Network (RON) [136], ZIP [156], Scale Transfer Detection Network (STDN) [321], RefineDet [308], StairNet [283], Path Aggregation Network (PANet) [174], Feature Pyramid Reconfiguration (FPR) [137], DetNet [164], Scale Aware Network (SAN) [133], Multiscale Location aware Kernel Representation (MLKP) [278] and M2Det [315], as shown in Table 7 and contrasted in Fig. 17.…”

Enhancing U-Net with Spatial-Channel Attention Gate for Abnormal Tissue Segmentation in Medical Imaging

“…Several recent attempts have been made using attention mechanisms to increase the capabilities of CNNs in various vision tasks, including classification [19], detection [20], segmentation [21], image captioning [22], and visual question answering [23]. Attention mechanisms guide the model to emphasize the most salient features, avoiding useless features that are beneficial for specific tasks.…”