Knowledge-aware Neuron Interpretation for Scene Classification

This repository contains the data and code for the paper "Knowledge-aware Neuron Interpretation for Scene Classification".

Data

All experimental results are conducted on the scenes datasets ADE20k with atomic concepts defined by annotations.

Experiments for Model Prediction Explanation

False Prediction

python Model_Prediction_Explanation/false_prediction.py 

True Prediction

python Model_Prediction_Explanation/true_prediction.py

Experiments for Compositional Explanation

Using Concept Filtering

The complete results is at: Compositional_Explanation/using_concept_filtering/results-TransE-Dismult-TransD-ProjE_CLIP.csv.

Experiments for Model Manipulation

Data and Code Preparation

• Replace "/site-packages/torchvision/models/resnet.py" in your environment with "Model_Manipulation/replace_code/resnet.py"

Disabling Positive Neurons

python Model_Manipulation/model_manipulation/run_disable_pos_neurons.py

Disabling Negative Neurons

python Model_Manipulation/model_manipulation/run_disable_nega_neurons.py

Dependencies

• pandas==0.25.3
• tqdm==4.30.0
• imageio==2.6.1
• scipy==1.4.1
• matplotlib==3.1.3
• Pillow==7.0.0
• seaborn==0.9.0
• scikit-image==0.16.2
• pyparsing==2.4.6
• pyeda==0.28.0
• pycocotools==2.0