Skip to content

Image Captioning with Sparsed Transformer and Persistent Memory

Notifications You must be signed in to change notification settings

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

Releases

No releases published

Packages

No packages published

Languages