Single Pass Unified Object Detection With Spatial Grid Regression

via “single-pass unified object detection with spatial grid regression”

* 🏆 2017: [Attention is All you Need (Transformer)](https://proceedings.neurips.cc/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html)

Unique: Pioneered the single-stage detection paradigm by formulating object detection as a direct spatial regression problem on a grid, eliminating the region proposal generation stage (RPN) used by two-stage detectors. Uses a unified loss function jointly optimizing bounding box regression (L2 loss) and class prediction (cross-entropy) across all grid cells in a single forward pass through a fully-convolutional architecture.