Skip to content

Latest commit

 

History

History
39 lines (25 loc) · 937 Bytes

README.md

File metadata and controls

39 lines (25 loc) · 937 Bytes

Pointly-Supervised Instance Segmentation

implement with mmdetection

label of sample points

requirement

mmdet

performance

Backbone Style Lr schd mask AP
R-50 pytorch 3x 36.1

usage

train

./dist_train.sh

display labels

python point_label_display.py configs/official/point_sup_r50_fpn_1x_coco.py --show-dir ./output

please follow official mmdetection instructions

version

  • offitial: modified from detectron2 PointSup

  • own: implementation from scratch with some features:

    • sample sites in batch
    • provide sample sites visual info
    • extend relative import for better decoupling with origin mmdetection project (without modifying origin package code)