search

Serenos / LWSIS

An official implementation of AAAI2023 paper "LWSIS: LiDAR-guided Weakly Supervised Instance Segmentation for Autonomous Driving"
Other
15 stars 0 forks source link

About Waymo experiment #3

Closed jiangxb98 closed 1 year ago

jiangxb98 commented 1 year ago

How do you handle waymo instance annotations? I use waymo dataset to train the condinst and boxinst on mmdet2.x. But boxinst's mAP is higher then condisnt mAP.I suspect I mishandled the instance annotations.

Serenos commented 1 year ago

As a fully supervised segmetation network, CondInst is theoretically superior to BoxInst (weakly supervised), and this point has been proved in the paper. I trained with the official BoxInst repository, which also provides an implementation of CondInst.

For the processing of Waymo instance segmentation data, we also first converted it into COCO format, and then conducted training and evaluation. I have attached the conversion script in the email, which can be run by main.py. So hopefully that answers your questions.

jiangxb98 commented 1 year ago

收到