search

lhwcv / mlsd_pytorch

Pytorch implementation of "M-LSD: Towards Light-weight and Real-time Line Segment Detection"
Apache License 2.0
189 stars 37 forks source link

The model *pth is your trained? or convert from official tensorflow tflite? #1

Closed xuanyuyt closed 3 years ago

xuanyuyt commented 3 years ago

Hi~, such as title, I want to know the answer to determine if it is easy to reproduce. thank you!

lhwcv commented 3 years ago

@xuanyuyt Hi. converted from TFLite. I also plan to reproduce, when get a result, I'll push the trainging code.

smitheric95 commented 3 years ago

Hi @lhwcv, could you share some info on how you converted the TFLite model to a .pth file? I would like to do the same for an older version of PyTorch.

(Great work, by the way!)

xuanyuyt commented 3 years ago

@xuanyuyt Hi. converted from TFLite. I also plan to reproduce, when get a result, I'll push the trainging code.

Thank you for your reply !

lhwcv commented 3 years ago

@smitheric95 Hi, thank you first. I don't found an ease-to-use tool convert the model from tflite to Pytorch. So wirte a network strictly according TFlite one, and use tflite + numpy to get the weight can be a choice.

Already push convert demo, you can see in convert/

smitheric95 commented 3 years ago

@lhwcv, I was able to successfully convert the TFLite model to .pth using your notebook in convert/.

Thank you very much!!