README_en.md

resnext101_32x8d_wsl

Module Name	resnext101_32x8d_wsl
Category	image classification
Network	ResNeXt_wsl
Dataset	ImageNet-2012
Fine-tuning supported or not	No
Module Size	317MB
Latest update date	-
Data indicators	-

I.Basic Information

• Module Introduction

  • The scale of dataset annotated by people is close to limit, researchers in Facebook adopt a new method of transfer learning to train the network. They use hashtag to annotate images, and trained on billions of social images, then transfer to weakly supervised learning. The top-1 accuracy of ResNeXt101_32x8d_wsl on ImageNet reaches 82.55%. This module is based on ResNeXt101_32x8d_wsl, and can predict an image of size 2242243.

II.Installation

• 1、Environmental Dependence

  • paddlepaddle >= 1.6.0

  • paddlehub >= 1.0.0 | How to install PaddleHub

• 2、Installation

  • $ hub install resnext101_32x8d_wsl

  • In case of any problems during installation, please refer to: Windows_Quickstart | Linux_Quickstart | Mac_Quickstart

III.Module API Prediction

• 1、Command line Prediction

  • $ hub run resnext101_32x8d_wsl --input_path "/PATH/TO/IMAGE"

  • If you want to call the Hub module through the command line, please refer to: PaddleHub Command Line Instruction

• 2、Prediction Code Example

  • import paddlehub as hub
    import cv2
    
    classifier = hub.Module(name="resnext101_32x8d_wsl")
    test_img_path = "/PATH/TO/IMAGE"
    input_dict = {"image": [test_img_path]}
    result = classifier.classification(data=input_dict)

• 3、API

  • def classification(data)

    • classification API.

    • Parameters

      • data (dict): key is "image", value is a list of image paths

    • Return

      • result(list[dict]): classication results, each element in the list is dict, key is the label name, and value is the corresponding probability

IV.Release Note

• 1.0.0

  First release

  • $ hub install resnext101_32x8d_wsl==1.0.0