ABSTRACT
Chen, S.D. and Liu, Y., 2020. Migration learning based on computer vision and its application in ocean image processing. In: Guido Aldana, P.A. and Kantamaneni, K. (eds.), Advances in Water Resources, Coastal Management, and Marine Science Technology. Journal of Coastal Research, Special Issue No. 104, pp. 281–285. Coconut Creek (Florida), ISSN 0749-0208.
Transfer learning is commonly used in computer vision, we want to apply it on visual question answering. We build up a QA pre-trained model to extract QA features from different source datasets and finally find the optimal pre-training for next experiments. We select LSTM model as our QA pre-train model since LSTM could fit different length of questions and has the same structure as our baseline model. We also do experiments on VQA baseline with LSTM+CNN model and pre-train + CNN model on different scales of datasets to find out the effectiveness of transfer learning. We figure out that baseline + pre-train models with different features have different influence. At last, we build up a VQA model with attention mechanism to combine QA pre-train features and image pre-train features, and studied its application in ocean image processing.
