PaddlePaddle / PaddleOCR

Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)
Apache License 2.0
39.34k stars 7.35k forks source link

EastRandomCropData的size如何配置为1080P #12055

Closed yaojun10086 closed 3 weeks ago

yaojun10086 commented 3 weeks ago

EastRandomCropData的size如何配置为1080P(1920*1080),修改完yml后执行检测训练报错ValueError: (InvalidArgument) Broadcast dimension mismatch. Operands could not be broadcast together with the shape of X = [8, 96, 135, 240] and the shape of Y = [8, 96, 136, 240]. Received [135] in X is not equal to [136] in Y at i:2. 错误异常提示在\tools\program.py", line 307, in train preds = model(images)

UserWangZz commented 3 weeks ago

你好,这是因为模型对图片下采样出现的问题,19201080的图像下采样8倍会变成 240135的特征图,但是与另一个数据shape不一致导致的报错,需要您排查一下Y张量和X张量