Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution

In natural images, information is conveyed at different frequencies where higher frequencies are usually encoded with fine details and lower frequencies are usually encoded with global structures. Similarly, the output feature maps of a convolution layer can also be seen as a mixture of information at different frequencies... (read more)

PDF Abstract ICCV 2019 PDF ICCV 2019 Abstract
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Image Classification ImageNet Oct-ResNet-152 (SE) Top 1 Accuracy 82.9% # 116
Top 5 Accuracy 96.3% # 58
Number of params 67M # 68
Action Classification Kinetics-400 Oct-I3D + NL Vid acc@1 75.7 # 57

Methods used in the Paper


METHOD TYPE
ResNeXt Block
Skip Connection Blocks
Depthwise Convolution
Convolutions
Pointwise Convolution
Convolutions
Depthwise Separable Convolution
Convolutions
Octave Convolution
Convolutions
Sigmoid Activation
Activation Functions
Dense Connections
Feedforward Networks
Squeeze-and-Excitation Block
Image Model Blocks
Grouped Convolution
Convolutions
Residual Connection
Skip Connections
ReLU
Activation Functions
Batch Normalization
Normalization
ResNeXt
Convolutional Neural Networks
Cosine Annealing
Learning Rate Schedules
SGD
Stochastic Optimization
Mixup
Image Data Augmentation
Label Smoothing
Regularization
Softmax
Output Functions
Average Pooling
Pooling Operations
1x1 Convolution
Convolutions
Inverted Residual Block
Skip Connection Blocks
MobileNetV2
Image Models
Bottleneck Residual Block
Skip Connection Blocks
Global Average Pooling
Pooling Operations
Residual Block
Skip Connection Blocks
Kaiming Initialization
Initialization
Max Pooling
Pooling Operations
ResNet
Convolutional Neural Networks