krasserm / fairseq-image-captioning

Transformer-based image captioning extension for pytorch/fairseq
Apache License 2.0
312 stars 55 forks source link

Bounding box encoding with sin/cos functions #11

Open krasserm opened 4 years ago

krasserm commented 4 years ago

Inspired by coordinate encoding done in https://arxiv.org/abs/2003.08934.