some problem when operating

[jiexiong2016]

some problem when operating

[dongdingcheng]

[jiexiong2016]

Hi,
Seems to be a loading error due to the path of GCNv2 model. The error should be be fixed if you run './rgbd_gcn' in folder GCN2 as the "run.sh" we provided. Alternatively, you could also provide the model path by using the environment variable "GCN_PATH" :

GCNv2_SLAM/src/GCNextractor.cc

Line 216 in 75cb4e6

const char *net_fn = getenv("GCN_PATH");

Besides, the vocabulary for GCNv2 should be "GCNvoc.bin".

[dongdingcheng]

@jiexiong2016 Thank you for answering my questions. I having figured out the problem. Its something wrong about the CUDA which my version is CUDA 9.0. Then I upgrade to the CUDA 9.2. I can run the operation. It is an excellent work!!!
And then, I wonder how to train my own vocabulary with my own network. Could you tell me some details?

[jiexiong2016]

Glad you solve the problem. Have just committed an example under GCNv2: GCNv2_SLAM/GCN2/create_voc.cc
The path to model and dataset should be modified to your own ones.

GCNv2_SLAM/GCN2/create_voc.cc

Line 128 in 13fadb0

glob("/home/t/Workspace/GCNV2/SUN3D/mono1/*.png", fn, false);

[dongdingcheng]

Thank you, I will try my own model~

[chaiein]

Hi,
Seems to be a loading error due to the path of GCNv2 model. The error should be be fixed if you run './rgbd_gcn' in folder GCN2 as the "run.sh" we provided. Alternatively, you could also provide the model path by using the environment variable "GCN_PATH" :

GCNv2_SLAM/src/GCNextractor.cc

Line 216 in 75cb4e6

const char *net_fn = getenv("GCN_PATH");

Besides, the vocabulary for GCNv2 should be "GCNvoc.bin".
Path to vocabulary should be GCNvoc.bin or ORBvoc.txt?
Program breaks at "module = torch::jit::load(net_fn); "
net_fn value is gcn2.pt
How to solve this issue

[jiexiong2016]

It shouldn't have loading error if you run the "run.sh" from GCN2 folder. I can only suggest to check the compilation and installation of pytorch (and its dependencies) since the problem is likely related to these.

[match08]

Glad you solve the problem. Have just committed an example under GCNv2: GCNv2_SLAM/GCN2/create_voc.cc
The path to model and dataset should be modified to your own ones.

GCNv2_SLAM/GCN2/create_voc.cc

Line 128 in 13fadb0

glob("/home/t/Workspace/GCNV2/SUN3D/mono1/*.png", fn, false);

no “saveToBinaryFile” function

[jiexiong2016]

Have added the function

[cumtchenchang]

Please forgive me not to open a new issue, my error is very related to this package. When I use create_voc.cc to train vocabulary, I encountered an error:
/chenchang1/data/CityCentre_test/0010.jpg
... Extraction done!
Creating vocabulary. May take some time ...
Segmentation fault (core dumped)
There are only ten images in my folder, this error is about "voc.create(features);"
Can you solve this problem? Thank you, sincerely.