Toybrick

标题: Zero-Copy API处理动态Shape输入问题 [打印本页]

作者: 萌虎龟来    时间: 2024-2-23 16:46
标题: Zero-Copy API处理动态Shape输入问题
我的模型为可变Shape输入,动态输入信息查询如下:
index=0, name=input, shape_number=2, range=[[1, 255, 255, 3], [1, 127, 127, 3]], fmt=NHWC

问题:
第一步  rknn_set_input_shapes 为[1, 127, 127, 3]  模型推理成功
第二步  rknn_set_input_shapes 为[1, 255, 255, 3]  模型推理成功
第三部  rknn_set_input_shapes 为[1, 127, 127, 3]  再次设置为127时,报错如下:
E RKNN: [07:54:37.150] rknn_set_input_shapes error, input name = input supported input shapes are as follows:
E RKNN: [07:54:37.150]         shape = [1,3,255,255], layout = NHWC
E RKNN: [07:54:37.150]         shape = [1,3,127,127], layout = NHWC
E RKNN: [07:54:37.150] while get rknn_tensor_attr[0].name = input, dims = [1, 127, 127, 3], fmt = NHWC


通用API则不会出现这个问题,需要重置哪些信息来解决这个错误???





欢迎光临 Toybrick (https://t.rock-chips.com/) Powered by Discuz! X3.3