Active Learning Supervised Cross-Modal Retrieval (ASCMR)

PyTorch implementation of ASCMR.

The code includes the implementations of all the baselines presented in the paper. Parts of the code are borrowed from https://github.com/JordanAsh/badge.

Setup

The dependencies are in requirements.txt. Python=3.8.3 is recommended for the installation of the environment.

Datasets

You can download the features of the datasets from:

• MIRFlickr, OneDrive, BaiduPan(password: b04z)
• NUS-WIDE (top-21 concepts), BaiduPan(password: tjvo)
• MS-COCO, BaiduPan(password: 5uvp)

Implementation

• The supervised cross-modal retrieval model implementations are followed by GNN4CMR.
• The active learning processes are followed by ALFA-Mix.

Training

python main.py \
        --data_name NUS-WIDE-TC21 --data_dir data/NUS-WIDE-TC21/ --log_dir log_only_label_retrain_sim_gcn --n_init_lb 500 \
        --n_query 500 --n_round 10 --learning_rate 0.00005 --n_epoch 1000 --model dagnn \
        --strategy $STRATEGY --alpha_opt --cuda_visible_devices 2 --map_threshold 0.99