method: ESRNet2023-11-29

Authors: Huageng Fan

Description: Our model utilizes ResNet50 as the backbone network. In our work, the training process is divided into pretraining and fine-tuning stages. For the ICDAR19-ArT dataset, we conduct pretraining on SynthText150k, MLT, LSVT, and ICDAR19-ArT for 500k iterations and fine-tune on the ICDAR19-ArT dataset for 50k iterations.