1 paper · 1 filter
Somak Aditya, Rudra Saha, Yezhou Yang +1
For tasks involving language and vision, the current state-of-the-art methods tend not to leverage any additional information that might be present to gather relevant (commonsense)…