It seems much more efficient than other conventional backbone models. And there is a well-maintained repo in github: https://github.com/lukemelas/EfficientNet-PyTorch
We are working are some efficient backbones in the team and planning to add them to Detectron2 soon.
Cool, that will be fantastic. Can't wait to see this.
@wanyenlo Any update on this? Thanks!
@wanyenlo any update on this? thanks.
@xiongzubiao @DanielRHarris @zimenglan-sysu-512
I create efficient backbone networks called VoVNet for detectron2.
vovnet_for_detectron2
@wanyenlo Hi, are there any updates on this? Did your team start with EfficientDet migration from TF to PyTorch / Detectron2?
We have released some efficient classification models (including EN) in the pycls model zoo: https://github.com/facebookresearch/pycls. However, porting these models to detectron2 still requires some manual work.
We may eventually make it easy to port models trained in pycls to detectron2 as backbones, but there is no immediate plan for this. There are no plans to add EfficientDet at the moment.
Most helpful comment
We are working are some efficient backbones in the team and planning to add them to Detectron2 soon.