Granular Concept Circuit (GCC)

Dahee Kwon* · Sehyun Lee* · Jaesik Choi

This is the official implementation of Granular Concept Circuits: Toward a Fine-Grained Circuit Discovery for Concept Representations, published in ICCV 2025.

Abstract

Deep vision models have achieved remarkable classification performance by leveraging a hierarchical architecture in which human-interpretable concepts emerge through the composition of individual neurons across layers. Given the distributed nature of representations, pinpointing where specific concepts are encoded within a model remains a crucial yet challenging task in computer vision. In this paper, we introduce an effective circuit discovery method, called Granular Concept Circuit (GCC), in which each circuit represents a concept relevant to a given query. To construct each circuit, our method iteratively assesses inter-neuron connectivity, focusing on both functional dependencies and semantic alignment. By automatically discovering multiple circuits, each capturing specific concepts within that query, our approach offers a profound, concept-wise interpretation of models and is the first to identify circuits tied to specific visual concepts at a fine-grained level.

Setup

1. To create the conda environment needed to run the code, run the following command:

conda env create -f environment.yaml
conda activate GCC

2. (Optional) If you use older GPU drivers (i.e., CUDA 11.8), please try this version of PyTorch.

pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu118

Now, you are ready to explore deep vision models with GCC.

Usage

1. Identify root nodes: We designate a node as a root node if its activation for a query sample falls within the top 1% across all samples. To identify such cases, we extract the top 1% highest activation samples for each node.

• Make sure to download the corresponding dataset. Here, we use the ImageNet validation dataset.
• Modify the output directory path according to your setup.

python3 get_highly_activated_samples.py --model_name resnet50 --dataset imagenet --n_samples 500 --out_dir /GCC/results/

2. Extract Granular Concept Circuits: Run the following script to extract circuits for a given query.

bash scripts/selected_resnet50.sh

3. Visualization (will be provided soon)