search

chenjianqu / RAFT_CPP

The optical flow algorithm RAFT implemented with C++(Libtorch+TensorRT)
MIT License
40 stars 4 forks source link

NotImplementError #1

Open chenxingxin-star opened 2 years ago

chenxingxin-star commented 2 years ago

hi, thank you for your project! may I ask if you modify the original RAFT Network in order to convert it to onnx ? because I found the input is fmap1, fmap2 = self.fnet([image1, image2],True), but the origin RAFT only input two images without the bool element 'True'. Following the Readme.md, I just get the error below: 屏幕截图 2022-02-02 17_33_43

could you give me some help? if you change the original RAFT code ,could you share what you have changed?

chenjianqu commented 2 years ago

I'm sorry, I missed a few notes. You should use my slightly modified RAFT project, that is https://github.com/chenjianqu/RAFT

chenxingxin-star commented 2 years ago

I'm sorry, I missed a few notes. You should use my slightly modified RAFT project, that is https://github.com/chenjianqu/RAFT

Thank you so much for your quick reply!

chenxingxin-star commented 2 years ago

when compile the code , I get these two error: first: 屏幕截图 2022-02-02 23_17_15 second: 屏幕截图 2022-02-02 23_16_46

my tensorRT version is tensorRT8, and my g++ version is g++-10. so why I get the errors when compile? thanks

chenjianqu commented 2 years ago

The first issue seems to be with the TensorRT version, I'm using TensorRT8.0.1 and there is no problem. The second problem is with the GCC version, which requires GCC to support at least C++17.

chenjianqu commented 2 years ago

Or, alternatively, you can modify the code slightly to make the project work.

Cai-RS commented 8 months ago

Or, alternatively, you can modify the code slightly to make the project work.

你好，感谢你的工作！你说你对原RAFT网络进行了一些小修改，能否说一下你具体修改了什么？修改了之后你是重新训练了网络，还是想办法用了原作者提供的权重呢？