Lin Manhui

PaddleOCR-VL has a minimum requirement for the compute capacity of GPUs. Please check the documentation for more details.

We are working on this. This will be fixed within this week.

目前交流群已经不活跃啦 不过issue里会有解答 如果有问题建议可以提issue

1. 数据分布之间可能存在gap，建议可以在公开数据集上做预训练，再在无人机数据上微调； 2. 这些问题比较适合通过后处理解决，可以试试下面三个后处理函数： https://github.com/PaddlePaddle/PaddleRS/blob/f62b46fd7fffc7429565c69f7ee4b0984efe64f2/paddlers/utils/postprocs/crf.py https://github.com/PaddlePaddle/PaddleRS/blob/f62b46fd7fffc7429565c69f7ee4b0984efe64f2/paddlers/utils/postprocs/mrf.py https://github.com/PaddlePaddle/PaddleRS/blob/f62b46fd7fffc7429565c69f7ee4b0984efe64f2/paddlers/utils/postprocs/regularization.py