CaoWGG / CenterNet-CondInst

Instance Segmentation based on CenterNet and CondInst

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

CenterNet-CondInst

CenterNet: Objects as Points + CondInst: Conditional Convolutions for Instance Segmentation

Installation

Please refer to CnterNet INSTALL.md for installation instructions.

Training

## note : seg_weight default setting is 1. You can set it to other value to get better performance.
cd src
python main.py ctseg --exp_id coco_dla_1x --batch_size 20 --master_batch 9 --lr 1.25e-4 --gpus 0,1 --num_workers 4

Eval

## not support flip test and multi scale test
cd src
python test.py ctseg --exp_id coco_dla_1x --keep_res --resume

Visualization

cd src
python demo.py ctseg --exp_id coco_dla_1x --keep_res --resume --demo ../data/coco/val2017

Result

type AP AP50 AP75 APs APm APl
box 0.358 0.540 0.384 0.154 0.391 0.535
mask 0.306 0.493 0.317 0.100 0.341 0.490

backbone=dla_34, batch=32

Reference

  1. CenterNet
  2. CondInst

About

Instance Segmentation based on CenterNet and CondInst

License:MIT License


Languages

Language:Python 66.7%Language:Cuda 12.0%Language:C++ 10.7%Language:C 10.3%Language:Shell 0.3%Language:Makefile 0.0%