How to optimize images

[tvone]

By default ppocrv5_server_rec model has rec_img_shape as [3,48,320], can I change it to [3,48,640]? What settings for rec_img_shape does the pre-generated model support?

[liuhongen1234567]

Hello, the parameters can be changed to [3,48,640], but during training with ppocrv5_server_rec, the scaling ranges between [[320, 32], [320, 48], [320, 64]]. Changing it to 640 may affect the final recognition accuracy.
The configuration file for ppocrv5_server_rec is as follows:

PaddleOCR/configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml

Line 101 in b34d8fa

scales: [[320, 32], [320, 48], [320, 64]]

[tvone]

Hello, the parameters can be changed to [3,48,640], but during training with ppocrv5_server_rec, the scaling ranges between [[320, 32], [320, 48], [320, 64]]. Changing it to 640 may affect the final recognition accuracy. The configuration file for ppocrv5_server_rec is as follows:

PaddleOCR/configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml

Line 101 in b34d8fa

scales: [[320, 32], [320, 48], [320, 64]]

Thank you