Deep-Cross-Modal-Projection-Learning-for-Image-Text-Matching

Deep-Cross-Modal-Projection-Learning-for-Image-Text-Matching copied to clipboard

How to reproduce the performence on CUHK-PEDES?

2

I use python 3.6.4, Pytorch1.0.0 & torchvision 0.2.1, scipy 1.2.1. The results in paper 'Deep Cross-Modal Pojection Learning for Image-Text Matching' on CUHK-PEDES are:{top- 1 = 49.37%，top-10 = 79.27%}, but...

liyingGao

File "/home/dalian/xuxu/Image-Text-Matching/train.py", line 96, in main ema = EMA(args.ema_decay) NameError: name 'EMA' is not defined

liuqcm

Why not ResNet

4

Notice that the results in paper 'Deep Cross-Modal Pojection Learning for Image-Text Matching' are:{top- 1 = 49.37%，top-10 = 79.27%} while the results in this project are {top- 1 = 42.999%,top-10...

FangmingZhou

the argment "checkpoint_dir" should be "model_path"

FangmingZhou

FangmingZhou

Could you please give information regarding pre-processing of the data? Is there any requirement of downloading the raw data or pre-processed will work?

kajal15003

About the Performance

1

Hi, I appreciate your pytorch version of this work. I want to know if this code can reproduce the same performance/results as the paper reports (cmpm:rank1 44.02 cmpm+cmpc:rank1 49.37 on...

QUANRJ

Hello. Thank you for your excellent work! But I noticed a problem with CMPC Loss. In the paper, "the vector projection of image feature xi onto normalized text feature ̄...

ZhanYang-nwpu

KL散度是不是写反了？

1

原论文里面说pi是预测概率分布，qi是真实概率分布。如果是这样的话，那Li应该等于\sum{j=1}{n}qijlog(qij/pij)吧？

lizimo-n

nishibendanhaha