search

youngwanLEE / centermask2

[CVPR 2020] CenterMask : Real-time Anchor-Free Instance Segmentation
Other
772 stars 159 forks source link

Not getting good results with pretrained weights on CPU #80

Closed cansik closed 3 years ago

cansik commented 3 years ago

I tried to use the centermask2 on a CPU in WSL and after some forth and back it seems to run. But the results I get are really bad, I can not get it to recognize anything on the images (for example COCO_val2014_000000012639.jpg). Here is the command I use to inference it on an image (maybe interesting as well for others who struggled running this on a CPU):

python demo/demo.py --config-file "configs/centermask/centermask_V_39_eSE_FPN_ms_3x.yaml" --input COCO_val2014_000000012639.jpg --output output --confidence-threshold 0.1 --opts MODEL.DEVICE cpu

With a confidence-threshold set to 0.05 it starts to recognize something (but it's mostly uselsess). Do I miss something regarding the model or the input data?

Would be great if you could give me a hint on how to use the pre-trained model.

youngwanLEE commented 3 years ago

You should add this argument MODEL.WEIGHTS centermask2-V-39-eSE-FPN-ms-3x.pth.

cansik commented 3 years ago

Great! thank you very much!