Libra R-CNN: Towards Balanced Learning for Object Detection

Compared with model architectures, the training process, which is also crucial to the success of detectors, has received relatively less attention in object detection. In this work, we carefully revisit the standard training practice of detectors, and find that the detection performance is often limited by the imbalance during the training process, which generally consists in three levels - sample level, feature level, and objective level... (read more)

PDF Abstract CVPR 2019 PDF CVPR 2019 Abstract

Datasets


Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Object Detection COCO minival Libra R-CNN (ResNet-50 FPN) box AP 38.5 # 94
AP50 59.3 # 57
AP75 42.0 # 56
APS 22.9 # 49
APM 42.1 # 53
APL 50.5 # 58
Object Detection COCO test-dev Libra R-CNN (ResNeXt-101-FPN) box AP 43.0 # 99
AP50 64 # 75
AP75 47 # 82
APS 25.3 # 85
APM 45.6 # 94
APL 54.6 # 94

Methods used in the Paper


METHOD TYPE
Average Pooling
Pooling Operations
ResNeXt Block
Skip Connection Blocks
Non-Local Operation
Image Feature Extractors
Embedded Gaussian Affinity
Affinity Functions
Non-Local Block
Image Model Blocks
IoU-Balanced Sampling
Prioritized Sampling
Balanced L1 Loss
Loss Functions
Balanced Feature Pyramid
Feature Pyramid Blocks
Focal Loss
Loss Functions
FPN
Feature Extractors
Step Decay
Learning Rate Schedules
Libra R-CNN
Object Detection Models
Grouped Convolution
Convolutions
Bottleneck Residual Block
Skip Connection Blocks
Global Average Pooling
Pooling Operations
Residual Block
Skip Connection Blocks
Residual Connection
Skip Connections
ReLU
Activation Functions
Kaiming Initialization
Initialization
Max Pooling
Pooling Operations
1x1 Convolution
Convolutions
Batch Normalization
Normalization
RetinaNet
Object Detection Models
ResNeXt
Convolutional Neural Networks
Convolution
Convolutions