CornerNet-Lite: Efficient Keypoint Based Object Detection

18 Apr 2019 Hei Law Yun Teng Olga Russakovsky Jia Deng

Keypoint-based methods are a relatively new paradigm in object detection, eliminating the need for anchor boxes and offering a simplified detection framework. Keypoint-based CornerNet achieves state of the art accuracy among single-stage detectors... (read more)

PDF Abstract

Datasets


Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Real-Time Object Detection COCO CornerNet-Squeeze MAP 34.4 # 20
FPS 33 # 13
inference time (ms) 30 # 9
Object Detection COCO minival CornerNet-Saccade (Hourglass-104) box AP 41.4 # 65
APS 23.8 # 42
APM 43.5 # 48
APL 57.1 # 31
Object Detection COCO minival CornerNet-Saccade (Hourglass-54) box AP 42.6 # 54
APS 25.5 # 30
APM 44.3 # 41
APL 58.4 # 24
Object Detection COCO test-dev CornerNet-Squeeze box AP 34.4 # 156
Object Detection COCO test-dev CornerNet-Saccade (Hourglass-104, multi-scale) box AP 43.2 # 96
APS 24.4 # 93
APM 44.6 # 105
APL 57.3 # 66

Methods used in the Paper


METHOD TYPE
Depthwise Convolution
Convolutions
Pointwise Convolution
Convolutions
Residual Connection
Skip Connections
Convolution
Convolutions
Hourglass Module
Image Model Blocks
Corner Pooling
Pooling Operations
Stacked Hourglass Network
Pose Estimation Models
Depthwise Separable Convolution
Convolutions
Sigmoid Activation
Activation Functions
1x1 Convolution
Convolutions
Batch Normalization
Normalization
ReLU
Activation Functions
Soft-NMS
Proposal Filtering
Max Pooling
Pooling Operations
CornerNet-Squeeze Hourglass Module
Image Model Blocks
Depthwise Fire Module
Image Model Blocks
ColorJitter
Image Data Augmentation
Random Resized Crop
Image Data Augmentation
Random Horizontal Flip
Image Data Augmentation
Adam
Stochastic Optimization
CornerNet-Saccade
Object Detection Models
CornerNet-Squeeze
Object Detection Models
CornerNet-Squeeze Hourglass
Convolutional Neural Networks