yanfangcs / cyctr-pytorch Goto Github PK

Hi, congrats on your great work, I seem a little late to the party. Still, after looking around, I can't figure out how to make an inference about an unlabeled image (an image that doesn't contain any prior information about the label).

My question is how can I run inference on unlabel images? Cause there will be no support images for the unlabeled image. And what is the practical way to work around this if not implemented yet?

Hello, I encountered this issue while running your code:/site-packages/MultiScaleDeformableAttention-1.0-py3.6-linux-x86_64.egg/MultiScaleDeformableAttention.cpython-36m-x86_64-linux-gnu.so: undefined symbol: THPVariableClass

How to solve it

Hi, I couldn't find the trained model for coco benchmark so

I wonder you are able to share it.

Thank you

I am trying to get the same performance as in the paper but it seems better than the paper.

I tested on COCO-20 with resnet50.

Fold 0 ( Split 0) shows 0.4128 accuracy.
[2022-05-24 14:44:59,046 INFO test.py line 247 1101140] meanIoU---Val result: mIoU 0.4128.
I guess this is correct mIoU for split 0 ?

It's supposed to be 38.9 based on the paper.

Is there something I missed?
and I wonder what's the meaning of classes in config files.

Hi ! thank you for sharing your work.

Is there an easy way to visualize results with the pre-trained model ?