Skip to content

2014gaokao/ImageCaptioning

Repository files navigation

ImageCaptioning

Modification based on AoANet with higher score

Requirements

  • Python 3.6
  • Java 1.8
  • PyTorch 1.0
  • cider
  • coco-caption
  • tensorboardX

Model


image

Training

Prepare data

See details in data/README.md

Start training

# CUDA_VISIBLE_DEVICES=0 sh train.sh

Evaluation

# CUDA_VISIBLE_DEVICES=0 sh test.sh

Performance

method Bleu_1 Bleu_2 Bleu_3 Bleu_4 METEOR ROUGE_L CIDEr
AoANet 0.8054903453672397 0.6523038976984842 0.5096621263772566 0.39140307771618477 0.29011216375635934 0.5890369750273199 1.2892294296245852
Ours 0.8072861444505534 0.6546857633417253 0.5105614254570193 0.391872107934891 0.2892757681786928 0.5890655008118075 1.2896424046401054

About

Image Captioning with Sparsed Transformer and Persistent Memory

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages