search

stanfordnlp / mac-network

Implementation for the paper "Compositional Attention Networks for Machine Reasoning" (Hudson and Manning, ICLR 2018)
Apache License 2.0
492 stars 119 forks source link

Training on VQAv2 #16

Open TheShadow29 opened 5 years ago

TheShadow29 commented 5 years ago

I see that there is option to choose dataset to be VQA. Wanted to know if I could train it on VQAv2 and if so how?

dorarad commented 5 years ago

Hi yea definitely! The flag is --dataVer, you can set it either to 1 or 2, the default is actually 2, haven't tested it recently on 1. I made local test some time ago and it was working with for both with no problems so I strongly believe the online version will work for you smoothly but just in case I would like to validate that so I'll get back to you in a couple of days with the official scores I get for VQA 1 and 2!