search

rockchip-linux / rknn-toolkit

BSD 3-Clause "New" or "Revised" License
808 stars 173 forks source link

转换时间序列分类模型报错 #387

Open Magus4756 opened 1 year ago

Magus4756 commented 1 year ago

我想转换一个onnx模型，用于时间序列的分类。模型为shufflenet v2，所有conv、gap等都是1d的。输入为[batch, channel, point_num]的npy文件。 load时报警： W Warning: Axis may need to be adjusted according to original model shape. build时报错： E The mean_values length of input 1 must equal 1, such as "0 " (MEAN )! 请问该如何解决？

eRaul commented 1 year ago

config没填对。比如模型有1个输入，shape是[1, 1, 196, 768]这种，config中的 mean_values, std_values 可以参考下面这么写： mean_values=[[0]], std_values=[[1]]