Question about adversarial Matching

rowanz / r2c

Recognition to Cognition Networks (code for the model in "From Recognition to Cognition: Visual Commonsense Reasoning", CVPR 2019)

https://visualcommonsense.com

MIT License

466 stars 91 forks source link

Question about adversarial Matching #8

Closed yangshao closed 5 years ago

yangshao commented 5 years ago

In the paper, it's said that "each answer appears exactly four times in the dataset". I try to verify this. But I cannot get this conclusion for the train.jsonl and val.jsonl file. Can you explain more about this? Besides, does this mean all correct answers appears 4 times, or all answers candidates appears 4 times? Thanks.

rowanz commented 5 years ago

answered here https://groups.google.com/forum/#!topic/visualcommonsense/-oIn2QLK1uA :)