Captioning-on-Sport-Images

Project Description

Image captioning is the automatic generation of the textual description of an image. In our project, sport related images are captioned.

Dataset

• Sport related images are acquired from both existing dataset,flickr 30k and manual collection from google.
• There are at least four different captions for each image of 12 sport types.
• 70% (1135) of dataset is used as training and 30% (480) is used as testing data.

Methodology

• Preprocessing: resize all images into 100 x 100, vectorize the captions
• Encoder (Feature Extraction): ResNet50
• Decoder (Sequence Generation): LSTM
• Evaluation: Bilingual Evaluation Understudy(BLEU) score