Skip to content

Latest commit

 

History

History
46 lines (32 loc) · 1.76 KB

README.md

File metadata and controls

46 lines (32 loc) · 1.76 KB
Raw

CHAIR - CBM-Enabled Human-AI Collaboration for Image Retrieval [IJCAI-24, CV4Animals@CVPR-24]

Code for the paper "Are They the Same Picture? Adapting Concept Bottleneck Models for Human-AI Collaboration in Image Retrieval" accepted at IJCAI-24 Human-Centered AI track and CV4Animals@CVPR-24 workshop.

Link to Paper webpage: CHAIR

Link to Paper PDF: Coming soon

Setup

  1. Create .env file in the root directory with the following content:
WANDB_API_KEY=your_wandb_api_key
WANDB_PROJECT=your_wandb_project_name
WANDB_ENTITY=your_wandb_entity
  1. Install the required packages:
pip install -r requirements.txt

Data

Follow the instructions for CUB from ConceptBottleneck and add the root directory to data_dir in the config file.

For CelebA, follow the instructions https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html. Note, the PyTorch vision download will most likely not work, so you will have to download the dataset manually and add the root directory to data_dir in the config file.

For AwA2, follow the instructions https://cvml.ist.ac.at/AwA2/. Note, the PyTorch vision download will most likely not work, so you will have to download the dataset manually.

OS Env

Set the following environment variables:

export AWA_DATA_DIR=/path/to/AwA2
export CUB_DATA_DIR=/path/to/CUB
export DATASET_DIR=/path/to/CelebA

Scripts for SLURM

bash scripts/run_train.sh scripts/run_retrieval.sh JOB_NAME SEED DATASET_NAME TRAIN_MODE

Training modes: Sequential or Joint Scripts: chair_retrieval.py or chair_stage_two_retrieval.py add to scripts/run_retrieval.sh