search

karpathy / neuraltalk2

Efficient Image Captioning code in Torch, runs on GPU
5.49k stars 1.26k forks source link

More Specific Captions - Train New Dataset? #83

Open dannybess opened 8 years ago

dannybess commented 8 years ago

Hi, I'm attempting to figure out how to get the captions generated by neuraltalk2 to be more specific / accurate. Would I have to retrain a whole new dataset? If so, is there anywhere where I could find one (trained or untrained)?

Thanks!

cjds commented 8 years ago

More specific how? They are training on COCO captions which are scene descriptions and aren't very specific. You could in theory caption all the images yourself and run from the pretrained checkpoint so the network would learn quickly