Segmentation models with pretrained backbones. PyTorch.
encoder_name="mobileone_s{0..4}"
).Thanks @VadimLevin, @kevinpl07, @Abd-elr4hman
encoder_name="mit_b0"
(or mit_b1..b5) to create a model.smp.metrics
module with different metrics based on confusion matrix, see docs
utils
module is going to be deprecated, if you still need it import it manually from segmentation_models_pytorch import utils
Thanks a lot for all contributors!
timm
: mobilenetv3 (#355) and gernet (#344)smp.losses
module (smp.utils.losses
would be deprecated in future versions)in_channels > 3
Thanks to @azkalot1 @JulienMaille @originlake @Kupchanski @loopdigga96 @zurk @nmerty @ludics @Vozf @markson14 and others!
Thanks to @azkalot1 for the new encoders and architecture!
timm-efficientnet*
And some small fixes.
Thanks @IlyaDobrynin @gavrin-s @lizmisha @suitre77 @thisisiron @phamquiluan and all other contributers!