A visual-linguistic model that uses GNNs for improving input and output representations. Project for EECE 571F.
- You can view the model configurations with all hyperparameters at /cfgs/refcoco.
- The dataloader is updated to perform edge feature construction. See lines 253 onwards in refcoco.py for more details.
- GNN is implemented here: gnn.py. An alternate GNN model is implemented in gnnV2.py.
- The model is updated to use the GNN and handle different GNN types. See lines 131 onwards in visual_linguistic_bert.py
Use the instructions provided in README-OLD.md.