Hi! We have experimented with vision features, but have not had great results with it. The experiments are in the appendix of the paper. We found that using vision features did not bring big improvements to the current model, but also cost a lot of resources to process.
Hi! We have experimented with vision features, but have not had great results with it. The experiments are in the appendix of the paper. We found that using vision features did not bring big improvements to the current model, but also cost a lot of resources to process.