Discover published sets by community

Real-time object detection system, single neural network predicts bounding boxes and class probabilities directly from full images, high speed with reasonable accuracy.

Method for detecting objects in images using a single deep neural network, performs well on various object sizes, balances speed and accuracy.

An advanced version of R-CNN, uses Region Proposal Networks (RPN) to generate potential bounding boxes in an image and to classify the objects, slower but more accurate than YOLO or SSD.

Improves upon the original R-CNN by sharing computation for different region proposals, uses ROI pooling to extract features, faster and more efficient than R-CNN.

Employs a focal loss function to address class imbalance during training, achieves a good balance between speed and accuracy, can detect objects at multiple scales.

Extension of Faster R-CNN with pixel-level segmentation, provides high-quality object instance segmentation, more computationally expensive.

Scalable and efficient object detection model, uses a compound scaling method, achieves higher accuracy with fewer parameters and FLOPS compared to previous models.

A category of methods that remove the need for predefined anchor boxes, detects object centers and directly regresses to bounding boxes, simplifies the detection pipeline.

Lightweight deep neural networks for mobile and embedded vision applications, use depthwise separable convolutions for efficient performance, suitable for real-time object detection on mobile devices.