GCNet

Why we included this project

GCNet is the official code behind the Global Context block, an attention module that gives detection and segmentation models a view of long-range dependencies without the heavy cost of full non-local networks. It runs on top of MMDetection, so if you already work in that PyTorch ecosystem you can read the configs, drop the GC block into your own detector, and skip the usual infrastructure work. The repo also tracks the paper's trajectory, from a Best Paper award at the ICCV 2019 Neural Architects Workshop to an extended TPAMI version, which helps when you're weighing whether the accuracy gain is worth the extra compute.