An Octave Convolution (OctConv) stores and process feature maps that vary spatially “slower” at a lower spatial resolution reducing both memory and computation cost. It takes in feature maps containing tensors of two frequencies one octave apart, and extracts information directly from the low-frequency maps without the need of decoding it back to the high-frequency. The motivation is that 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.

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

Latest Papers

PAPER DATE
Generalized Octave Convolutions for Learned Multi-Frequency Image Compression
Mohammad AkbariJie LiangJingning HanChengjie Tu
2020-02-24
Classification of LiDAR Data Combined Octave Convolution With Capsule Network
H. WuM. CaoA. WangM. Wang
2020-01-09
Multi-scale Octave Convolutions for Robust Speech Recognition
Joanna RownickaPeter BellSteve Renals
2019-10-31
Accurate Retinal Vessel Segmentation via Octave Convolution Neural Network
| Zhun FanJiajie MoBenzhang QiuWenji LiGuijie ZhuChong LiJianye HuYibiao RongXinjian Chen
2019-06-28
Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution
| Yunpeng ChenHaoqi FanBing XuZhicheng YanYannis KalantidisMarcus RohrbachShuicheng YanJiashi Feng
2019-04-10

Tasks

TASK PAPERS SHARE
Image Compression 1 14.29%
MS-SSIM 1 14.29%
Robust Speech Recognition 1 14.29%
Speech Recognition 1 14.29%
Retinal Vessel Segmentation 1 14.29%
Image Classification 1 14.29%
Video Recognition 1 14.29%

Components

COMPONENT TYPE
🤖 No Components Found You can add them if they exist; e.g. Mask R-CNN uses RoIAlign

Categories