thanks for your reply. the terminal command is&n

Much appreciated. I'll have a try as you said. <span clas

Interesting link to add a constant : <a href="https://github.com/NVIDIA/TensorRT/t

In node -1 (importClip): UNSUPPORTED_NODE: Assertion failed: inputs.at(2).is_weights() && "Clip max value must be an initializer!" about sparseinst_tensorrt HOT 7 CLOSED

leandro-svg commented on May 25, 2024

In node -1 (importClip): UNSUPPORTED_NODE: Assertion failed: inputs.at(2).is_weights() && "Clip max value must be an initializer!"

from sparseinst_tensorrt.

Comments (7)

leandro-svg commented on May 25, 2024

Dear WillianLee,
I'm glad you could use this repository.

Would you share with me your terminal command and the error log such that i could see where in the ONNX graph the error happens?

Also, I could find a useful link for this issue :
onnx/onnx-tensorrt#562

It says that the clip layer must have a maximum value in the clip node. You can set the max value up by modifying the ONNX graph directly. Btw, my personal ONNX graph does only have minimal value in the Clip layer.
If you haven't heard about ONNX graph, I suggest you to install netron, a tool to check your onnx graph.
To modify it, you could use : https://github.com/NVIDIA/TensorRT/tree/master/tools/onnx-graphsurgeon

from sparseinst_tensorrt.

willianLee commented on May 25, 2024

thanks  for your reply.  the terminal command is  'python build_engine.py', parameters are all default. tensorrt log is in the attachment.

…

------------------ 原始邮件 ------------------ 发件人: "leandro-svg/SparseInst_TensorRT" ***@***.***>; 发送时间: 2022年8月24日(星期三) 下午3:55 ***@***.***>; ***@***.******@***.***>; 主题: Re: [leandro-svg/SparseInst_TensorRT] In node -1 (importClip): UNSUPPORTED_NODE: Assertion failed: inputs.at(2).is_weights() && "Clip max value must be an initializer!" (Issue #1) Dear WillianLee, I'm glad you could use this repository. Would you share with me your terminal command and the error log such that i could see where in the ONNX graph the error happens? Also, I could find a useful link for this issue : onnx/onnx-tensorrt#562 It says that the clip layer must have a maximum value in the clip node. You can set the max value up by modifying the ONNX graph directly. Btw, my personal ONNX graph does only have minimal value in the Clip layer. If you haven't heard about ONNX graph, I suggest you to install netron, a tool to check your onnx graph. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

from sparseinst_tensorrt.

leandro-svg commented on May 25, 2024

Since i have never met this issue, it could be due to some different versions of TensorRT or ONNX. I am using ONNX version 1.9.0 and TensorRT version 8.0.1.6 on my jetson TX2.
Here is a link where which you can find different version of Clip node for different ONNX : https://github.com/onnx/onnx/blob/main/docs/Operators.md#Clip
. Anyway, try to set a maximum value for the Clip node woth GraphSurgeon (see Issue link above) and it should be solved. I hope.

For Clip-6 :
Attributes
max : float (default is (3.402823e+38))
Maximum value, above which element is replaced by max
min : float (default is (-3.402823e+38))
Minimum value, under which element is replaced by min

from sparseinst_tensorrt.

willianLee commented on May 25, 2024

Much appreciated. I'll have a try as you said.

…

------------------ 原始邮件 ------------------ 发件人: "leandro-svg/SparseInst_TensorRT" ***@***.***>; 发送时间: 2022年8月24日(星期三) 下午4:25 ***@***.***>; ***@***.******@***.***>; 主题: Re: [leandro-svg/SparseInst_TensorRT] In node -1 (importClip): UNSUPPORTED_NODE: Assertion failed: inputs.at(2).is_weights() && "Clip max value must be an initializer!" (Issue #1) Since i have never met this issue, it could be due to some different versions of TensorRT or ONNX. Here is a link where which you can find different version of Clip node for different ONNX : https://github.com/onnx/onnx/blob/main/docs/Changelog.md#Clip-6 . Anyway, try to set a maximum value for the Clip node woth GraphSurgeon (see Issue link above) and it should be solved. I hope. For Clip-6 : Attributes max : float (default is (3.402823e+38)) Maximum value, above which element is replaced by max min : float (default is (-3.402823e+38)) Minimum value, under which element is replaced by min — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

from sparseinst_tensorrt.

leandro-svg commented on May 25, 2024

I'll try this out on another GPU (Nvidia 2070 Max-Q) and i'll keep you up to date :)
Thank you again for reaching out and good luck.

from sparseinst_tensorrt.

leandro-svg commented on May 25, 2024

Interesting link to add a constant :
https://github.com/NVIDIA/TensorRT/tree/main/tools/onnx-graphsurgeon/examples/05_folding_constants

from sparseinst_tensorrt.

leandro-svg commented on May 25, 2024

Someone got this Clip issue and could solve it by using a docker image with CUDA 11.7, Pytorch 1.13 and TensorRT 8.2.5.
See comment from @LiZheng1997: hustvl/SparseInst#75

from sparseinst_tensorrt.

In node -1 (importClip): UNSUPPORTED_NODE: Assertion failed: inputs.at(2).is_weights() && "Clip max value must be an initializer!" about sparseinst_tensorrt HOT 7 CLOSED

Comments (7)

Related Issues (3)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent