Closed kyakuno closed 3 years ago
mit
@kyakuno The export of this model is trickier than I expected. It does not use a straightforward model under nn.Module
with the forward
method accepting a simple torch.Tensor
and as such, I have been refactoring the code a bit to get around this.
Moreover, even after exporting the ONNX model, I get the following error when running with ailia:
Would you have an idea in which cases a dead-lock could be caused?
@zhaochow Please implement --onnx option using onnx runtime for test. ailia SDK 1.2.6 will fix this issue.
@kyakuno Would you know where I can find some anime character images free to use as a sample? Also, there seem to be 2 models (CondInst and SOLOv2). I am finishing the export of SOLOv2 but should I also export the second one?
You only need to export one model. So you do not need export condinst model.
Please use some anime image download from https://unity-chan.com/contents/guideline/ . And please add this license. © Unity Technologies Japan/UCL
You only need to export one model. So you do not need export condinst model.
Understood!
I added the sample image and the image license in README.md. Please see the related PR (#337) for the changes.
https://github.com/zymk9/Yet-Another-Anime-Segmenter