Toybrick

标题: HELP! Receiving error: nodes must be less than 1024 [打印本页]

作者: lewis_brick 时间: 2019-7-4 15:19
标题: HELP! Receiving error: nodes must be less than 1024
[attach]366[/attach]
I received the error message above when transferring caffemodel. anyone can give me a hint?

作者: jefferyzhang 时间: 2019-7-23 10:03
I have reported this issue, please wait for response

作者: lewis_brick 时间: 2019-8-13 15:21

jefferyzhang 发表于 2019-7-23 10:03
I have reported this issue, please wait for response

thanks for your notice! if node number is limited, quite lot of network won't be able to be deployed on rk3399 pro
looking forward to the response

作者: lewis_brick 时间: 2019-9-2 10:39

jefferyzhang 发表于 2019-7-23 10:03
I have reported this issue, please wait for response

any feedback??

作者: yhc 时间: 2019-9-2 15:06
Hi, NPU driver limits the number of nodes must be less than 1024

RKNN-Toolkit will do node merging when the model is converted. For example, conv+bn+scale will be merged into a conv, which will reduce the number of nodes.So, can you modify the model to match the merge rule?

Currently, none of the models we tested exceeded this limit. Can this model be provided for us to test?

作者: lewis_brick 时间: 2019-9-3 11:15

yhc 发表于 2019-9-2 15:06
Hi, NPU driver limits the number of nodes must be less than 1024

RKNN-Toolkit will do node merging ...

The model I am using is hrnet for human keypoint detection. I transfer pytorch model to caffemodel, and then use rknn toolkit to obtain rknn model. during the rknn convertion, I encountered the error.

欢迎光临 Toybrick (https://t.rock-chips.com/)