we propose the Flowmind2digital method and the hdFlowmind dataset in this paper: Flowmind2digital is the first end-to-end recognition and conversion method for flowmind. Flowmind2digital uses a neural network architecture and keypoint detection technology to improve the overall recognition accuracy of the flowmind. Our experiments demonstrate the advantages of our method, whose accuracy on the hdFlowmind dataset is 87.3%, exceeding the previous SOTA work by 11.9%. At the same time, we also prove the effectiveness of our dataset. After pre-training our method and fine-tuning on Handwritten-diagram-dataset, the accuracy increased by 2.9%. Finally, we also discovered the importance of simple graphics for sketch recognition, that after adding simple data, the accuracy increased by 9.3%.
