search

mmaaz60 / EdgeNeXt

[CADL'22, ECCVW] Official repository of paper titled "EdgeNeXt: Efficiently Amalgamated CNN-Transformer Architecture for Mobile Vision Applications".
MIT License
351 stars 40 forks source link

EdgeNeXt for semantic segmentation #5

Closed lkdci closed 1 year ago

lkdci commented 2 years ago

Hi, Thanks for your work. Is there any plan to share your implementation for semantic segmentation with PascalVOC dataset? Perhaps instructions on how to build the modules or pth checkpoints / onnx models?

mmaaz60 commented 2 years ago

Hi @lkdci,

Thank you for your interest in our work. We do have plans to release segmentation codes and models, however it may take some time. Some extra details; we implemented our EdgeNeXt model in MobileViT for segmentation experiments. I am attaching the .pt checkpoints here in case it might be helpful for now. EdgeNeXt_DeepLabV3.zip

Thanks

lkdci commented 2 years ago

Hi @mmaaz60, thanks for the fast reply. So basically the backbone remains the same, and an ASPP module is used as Decoder?

mmaaz60 commented 2 years ago

Hi @mmaaz60, thanks for the fast reply. So basically the backbone remains the same, and an ASPP module is used as Decoder?

Hi @lkdci, yes you are right. Thanks

AndPuQing commented 2 years ago

@mmaaz60 Hi, Can give more details? I am trying to use edgenext for semantic segmentation of bdd100k, but after training 160k, the effect is not satisfactory, do you use any auxiliary head?

mmaaz60 commented 2 years ago

Hi @AndPuQing,

Apologies for the late reply. We do not use any auxiliary head. Did you try tuning your learning rate?