This model has been pushed to the Hub using the
PytorchModelHubMixin integration:
This is a decoder of image captioning model.
The image will be first preprocessed and resized to (224, 224) and then passed to ViT_b_32(with no classification layer), and then this will output
(N, 768). Then this will be repeated 32(max_length) times and will be passed to K, V to CrossMultiHeadAttention block in decoder. This model was trained with
Microsoft COCO2017 dataset and acheived 0.54 of masked_accuracy on validation set.
To use this model, first you need to download ViT_b_32 which will be used as encoder and download decoder from this repo.
