Skip to content

Latest commit

 

History

History
69 lines (42 loc) · 2.17 KB

octave-resnext-50-0.25.md

File metadata and controls

69 lines (42 loc) · 2.17 KB

octave-resnext-50-0.25

Use Case and High-Level Description

The octave-resnext-50-0.25 model is a modification of resnext-50 with Octave convolutions from Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution with alpha=0.25. Like the original model, this model is designed for image classification. For details about family of Octave Convolution models, check out the repository.

Example

Specification

Metric Value
Type Classification
GFLOPs 6.444
MParams 25.02
Source framework MXNet*

Accuracy

Performance

Input

A blob that consists of a single image of 1x3x224x224 in RGB order. Before passing the image blob into the network, subtract RGB mean values as follows: [124,117,104]. In addition, values must be divided by 0.0167.

Original Model

Image, name: data, shape: 1,3,224,224, format: B,C,H,W, where:

  • B - batch size
  • C - channel
  • H - height
  • W - width

Channel order is RGB. Mean values: [124,117,104], scale value: 59.880239521.

Converted Model

Image, name: data, shape: 1,3,224,224, format: B,C,H,W,where:

  • B - batch size
  • C - channel
  • H - height
  • W - width

Channel order is BGR.

Output

The model output for octave-resnext-50-0.125 is a typical object-classifier output for 1000 different classifications matching those in the ImageNet database.

Original Model

Object classifier according to ImageNet classes, name: prob, shape: 1,1000, output data format is B,C, where:

  • B - batch size
  • C - predicted probabilities for each class in [0, 1] range

Converted Model

Object classifier according to ImageNet classes, name: prob, shape: 1,1000, output data format is B,C, where:

  • B - batch size
  • C - predicted probabilities for each class in [0, 1] range

Legal Information

LICENSE