Implementation of Mask R-CNN in PyTorch

[Oktai15 (Oktai Tatanov)]

Hi there!

Does anybody have implementation of Mask R-CNN in PyTorch that has ability to fine-tuning on own dataset?

[zhanghaoinf (Hao Zhang)]

Hi, Oktai,

• For inference only, please refer to the below implementation (reproduce).
  https://github.com/ignacio-rocco/detectorch

• For training/fine-tuning, you can refer to this implementation, but no benchmark performance has been reported, meaning that it might has risks in reproducing paper’s results.
  https://github.com/roytseng-tw/mask-rcnn.pytorch

Best

[Oktai15 (Oktai Tatanov)]

@zhanghaoinf thanks a lot!

Did you try to train on own dataset with second model?

[zhanghaoinf (Hao Zhang)]

Not yet, I just saw a guy I stared follow the second implementation.

[zhanghaoinf (Hao Zhang)]

The author of the second implementation posts benchmark evaluation results, seems reasonable.

[lkhphuc (Phúc Lê)]

From the Facebook Research team, just released, support for training on own dataset.

facebookresearch/maskrcnn-benchmark