search

aim-uofa / AdelaiDet

AdelaiDet is an open source toolbox for multiple instance-level detection and recognition tasks.
https://git.io/AdelaiDet
Other
3.38k stars 648 forks source link

The CondInst is not good #189

Open cty-xiaobai opened 4 years ago

cty-xiaobai commented 4 years ago

000001 000001 000004CondInst Hi, I test the CondInst and MaskRcnn in detectron2. The first image is the original image, The second image is the result of maskrcnn, the third image is the result of CondInst .CondInst's results are not good. Do I need to adjust the code？ If anyone has a better solution, welcome to discuss it with me. Thanks a lot!

tianzhi0549 commented 4 years ago

@cty-xiaobai I think you should tune the threshold when testing.

cty-xiaobai commented 4 years ago

@cty-xiaobai I think you should tune the threshold when testing.

Thank you for your reply. Which threshold is and how to adjust？

tianzhi0549 commented 4 years ago

@cty-xiaobai Please lower the threshold here. https://github.com/aim-uofa/AdelaiDet/blob/d459a2fc9b23636274af2994761b67a888c952b2/demo/demo.py#L55

cty-xiaobai commented 4 years ago

@cty-xiaobai Please lower the threshold here.

https://github.com/aim-uofa/AdelaiDet/blob/d459a2fc9b23636274af2994761b67a888c952b2/demo/demo.py#L55

OK！thanks for your help！Decrease the threshold, the result is indeed better, but the threshold of detectron2 is also 0.5, I think it should be compared under the same threshold

tianzhi0549 commented 4 years ago

@cty-xiaobai please kindly note that different algorithms have different optimal thresholds.

cty-xiaobai commented 4 years ago

@cty-xiaobai please kindly note that different algorithms have different optimal thresholds.

Ok thanks for your help！

lxtGH commented 4 years ago

Hi! @tianzhi0549 what are the better thresholds for demo visualization?