Authors: yuchen su, tianlun zheng, yongkun du, yi gan, zhineng chen
Affiliation: Fudan University, Paddle OCR
Description: Our method is based on PAN, ResNet-50 pre-trained on ImageNet as our backbone. We only use the training images of ReST for training. For data augmentation, we apply random scale, random flip, random rotation and random crop on training images, and manually select difficult samples from training images for crop, color jitter, contrast jitter and occlusion data augmentation.
Wang W, Xie E, Song X, et al. Efficient and accurate arbitrary-shaped text detection with pixel aggregation network[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2019: 8440-8449.