SophiaSarah

Try --resize_or_crop=no_change

256x144 is workable. I guess 16:9 or 1:1 less than 256 is okay.

> 您好，可以给下运行命令。pretrain_model换为训练好的模型，checkpoint也换为训练好的模型。再试试看 `python3 tools/eval.py -c configs/det/det_PP-OCRv4_db.yml` 其中det_PP-OCRv4_db.yml内容如下 ``` Global: use_gpu: true use_xpu: false use_mlu: false epoch_num: 1200 log_smooth_window: 20 print_batch_step: 10 save_model_dir: ./output/ch_PP-OCRv4/ save_epoch_step: 1200 # evaluation is run every...