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 650 forks source link

The performance of Coninst and BoxInst on the Cityscapes dataset #333

Open sankin97 opened 3 years ago

sankin97 commented 3 years ago

Hello, I'm trying to train Coninst and BoxInst on the Cityscapes dataset, but i got a poor performance both of them. Can you give me some advices. Thank you.

tianzhi0549 commented 3 years ago

@sankin97 Could you please post your results here?

sankin97 commented 3 years ago

The boxinst results on cityscapes. image

sankin97 commented 3 years ago

@tianzhi0549 The CondInst result on Cityscapes(without coco pretrained) image

tianzhi0549 commented 3 years ago

@sankin97 You should use COCO pre-training.

sankin97 commented 3 years ago

@tianzhi0549 Thankyou, I'll trying the coco-pretraining. I am also confused about the results without pre-training, such results are normal or not？I have found a supplenmenray material for CondInst on Cityscapes，my result is worse than 36.9 maskAP you got. Can you share me with your detail settings on CondInst. And Is the BoxInst performance is bad due to the overlap between objects on Cityscapes? Can you give me some advices. Thankyou!

tianzhi0549 commented 3 years ago

@sankin97 The results do not seem normal. For both CondInst and BoxInst, we use the same settings as in https://github.com/facebookresearch/detectron2/blob/master/configs/Cityscapes/mask_rcnn_R_50_FPN.yaml.

sankin97 commented 3 years ago

@tianzhi0549 Thanks for your responses! I also follow the settings in https://github.com/facebookresearch/detectron2/blob/master/configs/Cityscapes/mask_rcnn_R_50_FPN.yaml in my experiments. I will do some experiments with the coco pre-training, then share with you, i look forward to your attention.

ZCMax commented 3 years ago

Can you offer me a pretrained model on Cityscapes?

luyao-cv commented 2 years ago

have you dealt with this problem?