Toybrick

Input channels limitation <= 128?

nicolausYes

新手上路

积分
26
发表于 2020-8-8 04:07:52    查看: 5288|回复: 0 | [复制链接]    打印 | 显示全部楼层
I've been told there is the limitation of <= 128 channels per input, could you please clarify?

I'm getting this message during the conversion using rknn-toolkit-1.3.3rc9.
W Channels(256) of input node: input_0> 128, mean/std values ​​will be set to default 0/1.
W Please do pre-processing manually before inference.
W Input channel 2 less then 3, please update NPU Driver to v1 .3.0 or later.
W Channels(512) of input node: input_3> 128, mean/std values ​​will be set to default 0/1.
W Please do pre-processing manually before inference.

After the conversion I have an *.rknn model which is probably not valid, because it's not running on a device.
Our model inputs are the following.
1) name: input_imgs
type: float32[1,12,128,256]
2) name: desire
type: float32[1,8]
3) name: traffic_convention
type: float32[1,2]
4) name: rnn_state
type: float32[1,512]
But converting Inception_ResNet_V2 or Inception_V4 that also have an input> 128 channels doesn't give any error.
name: input
type: float32[1,299,299,3]

回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则

产品中心 购买渠道 开源社区 Wiki教程 资料下载 关于Toybrick


快速回复 返回顶部 返回列表