Every Problem, Every Step, All in Focus: Learning to Solve Vision-Language Problems With Integrated Attention

A novel Solution Graph Attention Network (SGAN) approach that takes into account both intra-step and inter-step attention mechanisms, enabling a progressive construction of solutions by refining the dependencies between relevant problemsolving steps.

Citation

If you use our code or data, please cite our paper:

@article{xianyu:2024:sgan,
    Author         = {Xianyu Chen and Jinhui Yang and Shi Chen and Louis Wang and Ming Jiang and Qi Zhao},
    Title          = {Every Problem, Every Step, All In Focus: Learning to Solve Vision-Language Problems with Integrated Attention},
    journal        = {IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI)},
    Year           = {2024}
}