CSPNet: A New Backbone that can Enhance Learning Capability of CNN

27 Nov 2019Chien-Yao WangHong-Yuan Mark LiaoI-Hau YehYueh-Hua WuPing-Yang ChenJun-Wei Hsieh

Neural networks have enabled state-of-the-art approaches to achieve incredible results on computer vision tasks such as object detection. However, such success greatly relies on costly computation resources, which hinders people with cheap devices from appreciating the advanced technology... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
RESULT LEADERBOARD
Real-Time Object Detection COCO CSPResNeXt50-PANet-SPP MAP 33.4 # 13
FPS 58 # 3
Image Classification ImageNet CSPResNeXt-50 (Mish+Aug) Top 1 Accuracy 79.8% # 66
Top 5 Accuracy 95.2% # 38
Number of params 20.5M # 42

Methods used in the Paper


METHOD TYPE
ResNeXt
Convolutional Neural Networks
Batch Normalization
Normalization
Convolution
Convolutions
1x1 Convolution
Convolutions
Kaiming Initialization
Initialization
ReLU
Activation Functions
Residual Connection
Skip Connections
Global Average Pooling
Pooling Operations
Grouped Convolution
Convolutions
DenseNet
Convolutional Neural Networks
ResNet
Convolutional Neural Networks
Residual Block
Skip Connection Blocks
Bottleneck Residual Block
Skip Connection Blocks
Max Pooling
Pooling Operations
Polynomial Rate Decay
Learning Rate Schedules
SGD with Momentum
Stochastic Optimization
Weight Decay
Regularization
Step Decay
Learning Rate Schedules
Swish
Activation Functions
Spatial Attention Module
Image Model Blocks
Average Pooling
Pooling Operations
Sigmoid Activation
Activation Functions
CSPResNeXt
Convolutional Neural Networks
CSPResNeXt Block
Skip Connection Blocks
CSPDarknet53
Convolutional Neural Networks
Exact Fusion Model
Feature Pyramid Blocks
Adaptive Feature Pooling
Pooling Operations
RPN
Region Proposal
PAFPN
Feature Extractors
CSPPeleeNet
Convolutional Neural Networks
CSPDenseNet
Convolutional Neural Networks
Squeeze-and-Excitation Block
Image Model Blocks
Dense Connections
Feedforward Networks
PANet
Instance Segmentation Models
CSPDenseNet-Elastic
Convolutional Neural Networks
Softmax
Output Functions
RoIAlign
RoI Feature Extractors
Maxout
Activation Functions
Dropout
Regularization
Dense Block
Image Model Blocks
PeleeNet
Convolutional Neural Networks
Concatenated Skip Connection
Skip Connections
Bottom-up Path Augmentation
Feature Extractors
Darknet-53
Convolutional Neural Networks
ResNeXt Block
Skip Connection Blocks
DenseNet-Elastic
Convolutional Neural Networks
Leaky ReLU
Activation Functions
FPN
Feature Extractors
Two-Way Dense Layer
Skip Connection Blocks
Elastic Dense Block
Skip Connection Blocks
Mish
Activation Functions