/MGRIR

A Multi-granularity Graph-based Representation for Image Recognition

Primary LanguagePython

MGRIR

A Multi-granularity Graph-based Representation for Image Recognition

Graph prepared

For mnist datasets:

  1. Enter directory \MGRIR\graph\mnist, and replace the corresponding data path

  2. then run python to_h5.py

  3. If you want to filter edges, the code will in the find_boundary.py

    if weight >= 0: # you can set the threshold you want
    ...

The same is true for cifar data (skip)

Train the model

For mnist datasets:

  1. Enter directory \MGRIR, and replace the corresponding graph path and models

  2. Then run python main_train.py. The training model files will be stored in the ”checkpoints“ folder, and the training logs will be recorded in the ”runs“ folder.

  3. You can run the test.py to take the model named --checkpoint_path to test the accuracy of all training sets. Replace the graph path --data_dir

Visualization the graph

If you want to visualization the graph, more details can be found in find_boundary.py