multimodallearning / pytorch-mask-rcnn

Other
2.01k stars 557 forks source link

PyTorch 1.? #96

Closed FiReTiTi closed 4 years ago

FiReTiTi commented 4 years ago

Hi, is there an "easy" way to make it compatible with PyTorch 1 or later? PyTorch 0.3 seems to be really outdated now.

HelloDusty commented 4 years ago

I am face to the same problem but switch the NMS and roialign to your own, the models going to run maybe. Now I am having a try.

FiReTiTi commented 4 years ago

I gave up on this one and used the TorchVision version, must better!

shanniruo commented 4 years ago

Did you succeed? I also want to use the rolalign module , by the way, do you know why NMS module and roialign module should be compiled by CUDA code?

shanniruo commented 4 years ago

@HelloDusty Did you succeed? I also want to use the rolalign module , by the way, do you know why NMS module and roialign module should be compiled by CUDA code?

FiReTiTi commented 4 years ago

@shanniruo I gave up on this one and used the TorchVision version, must better!

shanniruo commented 4 years ago

@FiReTiTi Sorry, I didn't understand. What should I do ?

FiReTiTi commented 4 years ago

@shanniruo: As the version in the current library does not seem to be supported anymore, I've switched to the TorchVision implementation. There is now a Mask R-CNN implementation in PyTorch in the library TorchVision. It works really well and there is an example for pedestrian detection that can be easily modified for any applications.

shanniruo commented 4 years ago

@FiReTiTi tanks