This is the companion code for sparse-image-captioning.
This repo contains Soft-Attention model implemented in TensorFlow 1.9.
Please note that this code is for reference purposes; and although it still works, is largely outdated. For a more up-to-date implementation, see sparse-image-captioning.
Please follow the instructions at this repo.
Please refer to caption_COMIC/commands.sh for example training and inference commands.
for i in 0.8 0.9 0.95 0.975; do
python train_caption.py \
--name '' \
--rnn_name 'LSTM' \
--supermask_type 'regular' \
--supermask_sparsity_target ${i} \
--checkpoint_path "${CNN_CKPT}" \
--dataset_dir ${DSET:-''} \
--dataset_file_pattern ${DSET_PATTERN:-''} \
--log_root ${LOG_ROOT:-''} \
--gpu ${GPU} \
--run 1
done
for i in 0.8 0.9 0.95 0.975; do
python train_caption.py \
--name '' \
--rnn_name 'GRU' \
--supermask_type 'regular' \
--supermask_sparsity_target ${i} \
--checkpoint_path "${CNN_CKPT}" \
--dataset_dir ${DSET:-''} \
--dataset_file_pattern ${DSET_PATTERN:-''} \
--log_root ${LOG_ROOT:-''} \
--gpu ${GPU} \
--run 1
donedeclare -a dirs=(
"word_w256_LSTM_r512"
"word_w256_LSTM_r512_xu_REG_1.0e+02_init_5.0_L1_wg_5.0_ann_sps_0.90"
)
for dir in "${dirs[@]}"; do
for i in 1 2 3; do
python infer_v2.py \
--infer_checkpoints_dir "${LOG_ROOT}/mscoco_v3/${dir}/run_0${i}" \
--infer_set 'test' \
--save_attention_maps '' \
--dataset_dir ${DSET:-''} \
--gpu ${GPU}
done
doneThe checkpoints are available at this repo.
The project is open source under BSD-3 license (see the LICENSE file).
© 2019 Center of Image and Signal Processing, Faculty of Computer Science and Information Technology, University of Malaya.