datvuthanh / HybridNets

HybridNets: End-to-End Perception Network
MIT License
594 stars 121 forks source link

light weight backbone #25

Closed ForestWang closed 2 years ago

ForestWang commented 2 years ago

would you support some light weight backbones such as RepVGG which is GPU friendly?

xoiga123 commented 2 years ago

Sure thing! We hid it inside backbone.py but now we're bringing it to light (command-line). Use timm to replace your backbone. Try

python train.py -p bdd100k -bb repvgg_b0

This is basically what we do under the hood:

if backbone_name:
            self.encoder = timm.create_model(backbone_name, pretrained=True, features_only=True, out_indices=(2,3,4))  # P3,P4,P5
MartinPR307 commented 5 months ago

Hello, after switching to repvgg_b0, the following error occurs: image

MartinPR307 commented 5 months ago

@datvuthanh @xoiga123

MartinPR307 commented 5 months ago

@ForestWang Have you encountered the same problem?

HT24TH commented 3 weeks ago

I have the same problem @MartinPR307, any suggestion to solve that @datvuthanh @xoiga123. Thank you.