search

JingchunCheng / SegFlow

Demo code of the paper "SegFlow: Joint Learning for Video Object Segmentation and Optical Flow", in ICCV 2017
178 stars 41 forks source link

Layer conv1_flowS has unknown engine #7

Closed F2Wang closed 6 years ago

F2Wang commented 6 years ago

Hi Jingchun,

I have been stucked with this runtime error for a while, I have installed the version included in the repository and the path should be set up correctly(although I am not 100% sure). But whenever I attempt to run the demo I get this error:

I1206 12:39:48.895972 22295 layer_factory.hpp:77] Creating layer conv1_flowS F1206 12:39:48.895979 22295 layer_factory.cpp:69] Layer conv1_flowS has unknown engine.

Any advice will be greatly appreciated!

wasidennis commented 6 years ago

Hi,

Have you tried this: https://github.com/bqlabs/flownet/issues/1.

It looks like the problem is related to the CUDNN.

F2Wang commented 6 years ago

Thank you so much! The problem is gone and now I can run the code smoothly!

And another irrelevant question... I tried the demo code and It doesn't appear to do any on-line training on the first image and mask, is there anyway I can enable online training using the first ground truth segment?

wasidennis commented 6 years ago

No problem and good to know it is running!

Currently, the code in this repository does not contain the online training part. If you want to finetune the model on the first frame, please take a look at our another repository for the example: https://github.com/JingchunCheng/Seg-with-SPN. They have different models but the methodology for finetuning is the same.

F2Wang commented 6 years ago

Understood! Thank you so much for your quick response!