OpenGVLab / InternImage

[CVPR 2023 Highlight] InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions
https://arxiv.org/abs/2211.05778
MIT License
2.47k stars 231 forks source link

bad result from Image Demo Segmentation #306

Open YicanGgg opened 1 month ago

YicanGgg commented 1 month ago

我用ADE20K CUDA_VISIBLE_DEVICES=0 python image_demo.py \ datapics/demo.jpg \ configs/ade20k/upernet_internimage_t_512_160k_ade20k.py \ checkpoint_dir/seg/upernet_internimage_t_512_160k_ade20k.pth \ --palette ade20k

得到的结果非常好,

但是当我用cityscapes结果非常差, CUDA_VISIBLE_DEVICES=0 python image_demo.py \ datapics/demo.jpg \ configs/cityscapes/upernet_internimage_xl_512x1024_160k_mapillary2cityscapes.py \ checkpoint_dir/upernet_internimage_xl_512x1024_160k_mapillary2cityscapes.pth \ --palette cityscapes

demo

这是什么原因呢

YicanGgg commented 1 month ago

i changed the modell to segformer_internimage_xl_512x1024_160k_mapillary2cityscapes

works well

so maybe the problem is this model upernet_internimage_xl_512x1024_160k_mapillary2cityscapes