Global context aware RCNN for object detection

RoIPool/RoIAlign is an indispensable process for the typical two-stage object detection algorithm, it used to rescale proposal cropped from feature pyramid generate a fixed size map. However, these maps of local receptive fields will heavily lose global context information. To tackle this problem, we propose novel end-to-end trainable framework, called aware (GCA) RCNN, aiming at assisting neural network in strengthening spatial correlation between background and foreground by fusing The core component our GCA framework mechanism, which both attention strategies are extraction refinement, respectively. Specifically, leverage dense connection improve information flow different stages top-down FPN, further use mechanism refine each level pyramid. In end, also present lightweight version method, only slightly increases model complexity computational burden. Experimental results on COCO benchmark dataset demonstrate significant advantages approach.