Authors: Wudao, Liaoming
Description: DBnet++ with mobileone s3 backbone. In training phase, using 2w+ generate circle seal and rectangle seal images. In inference, multiscale the images with 320 size and 480 size, merging the results by picking the best confidence results.
M. Liao, Z. Zou, Z. Wan, C. Yao and X. Bai, "Real-Time Scene Text Detection With Differentiable Binarization and Adaptive Scale Fusion," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 1, pp. 919-931, 1 Jan. 2023, doi: 10.1109/TPAMI.2022.3155612.
Vasu, Pavan & Gabriel, James & Zhu, Jeff & Tuzel, Oncel. (2022). An Improved One millisecond Mobile Backbone. 10.48550/arXiv.2206.04040.
Source code 2