Readme

Model code implementation in Zero-shot Learning. Update ing.

Table of Contents

• Run
• Descriptions
  • DataProcess
  • Tools
  • ModelBuilder
• Related Work

Run

Simply run the file main.py:

$ python main.py

Descriptions

A brief introduction to each part of the code.

DataProcess

Load image dataset and get graph batch.

• graph_batch.py

  Using BFS to find most irrelated nodes and generate negative pairs.

• AdaptionFunc.py

  Fit the data to GoG model input form.

Tools

• GlobalGraph.py

  Generate Hierarchical Class Graph from Wordnet. (e.g., 285 edges and 286 nodes, where 200 are classes in CUB dataset.)

• LocalGraph.py

  Generate Local Attribute Graph for every class node.

ModelBuilder

• VisualModel.py

  Including Attention Net and Two Linear Layers.

• GogModel.py

Related Work

• [GoG] - GoG model
• [DAZLE] - DAZLE model

VisualGoG