Empirical Validation of Deep Learning Based on Image Captioning Models

Generating image captions automatically from the image is one of the complex problems in artificial intelligence. This study has empirically compared the different image captioning models. VGG16, InceptionV3, and ResNet50 are the three different convolutional neural network (CNN) models used in this study. The extracted features from the CNN models are passed in the unidirectional LSTM and attention LSTM (ALSTM) for the caption generation. The LSTM model used the beam search to generate the captions from the vocabulary. This study compared different CNN models over the two datasets, flicker8k and MS-COCO. We used METEOR, CIDEr, and BLEU scores for quantitative assessment to compare the different models.