raoyongming / HorNet

[NeurIPS 2022] HorNet: Efficient High-Order Spatial Interactions with Recursive Gated Convolutions

Home Page:https://hornet.ivg-research.xyz/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

How can I replicate the 55.0 mIoU results of HorNet-L-GF on ADE20K? The link of pretrained model seems to be valid

Linwei-Chen opened this issue · comments

What is the pretrained model? The link seems to be valid: model.
I used HorNet-L (GF)* | hornet_large_gf_img384 at IN22k-384, but it seems unable to achieve 55.0, reaching only around 54.5.

Hi @Linwei-Chen, we have updated the link. You are using the correct pre-trained model. The large model might be more sensitive to hyperparameters. You may try our pre-trained segmentation model and check the environment.

Thank you for your response! I will give it a try.