jiasenlu / NeuralBabyTalk

Pytorch code of for our CVPR 2018 paper "Neural Baby Talk"
https://arxiv.org/abs/1803.09845
MIT License
523 stars 123 forks source link

Ask for advice for the performance #42

Open xinfu607 opened 5 years ago

xinfu607 commented 5 years ago

Hello. I trained the model with the default settings (Standard Image Captioning), but I got a lower score than the reported score in you paper.

Here is my result: Bleu_1: 0.743 Bleu_4: 0.330 METOR:0.259 METOR:0.259 CIDEr: 1.021 SPICE: 0.193 (beam=3) The reported score in your paper. Bleu_1: 0.755 Bleu_4: 0.347 METOR:0.271 METOR:0.271 CIDEr: 1.072 SPICE: 0.201

Please give me advice for reproducing the results if convenient.