Home

Awesome

Prerequisites

Installation the environment

Please first refer to MattNet and KPRN (Knowledge-guided Pairwise Reconstruction Network for Weakly Supervised Referring Expression Grounding) to prepare related data. Then replace all files in this repository. "loaders" and "evals" should be placed in the "./lib/" and the main model file "model.py" should be placed in "./lib/layers/".

Pre-trained models

All pre-trained models and related data can be downloaded here. "mrcn_cmr_with_st.json","mrcn_cmr_with_st.pth","mrcn_cmr_with_st_critic.pth","mrcn_cmr_with_st_Re.pth" should be replaced in "./output/dataset_split/id_num/". "sent_extract.json" should be replaced in "./cache/sub_obj_wds/dataset_split/".

Generate the Triads

You can generate the triad file thourgh our previous work DTWREG.

Supervised Training

python ./tools/train_SL.py

Reinforcement Training

python ./tools/train_AC.py

Evaluation

python ./eval.py