Alexadlu / ACNet

ACNet: Attention Complementary Network for RGBD semantic segmentation

ACNET

This is the official implement for

ACNET: ATTENTION BASED NETWORK TO EXPLOIT COMPLEMENTARY FEATURES FOR RGBD SEMANTIC SEGMENTATION    
Xinxin Hu*, Kailun Yang, Lei Fei,  Kaiwei Wang
Accepted by IEEE ICIP 2019

Experiment result

we evaluate the mIoU of ACNet in SUN-RGBD and NYUDv2

	SUN-RGBD	NYUDv2
mIoU	48.1%	48.3%

How to use

This code is NYUDv2 implement.

Requestments

Python 3
Pytorch 0.4.1
TensorboardX
Tensorboard

Train the mode

python ACNet_train_V1_nyuv2.py --cuda -b 4

Evaluate the mode

The pretrained model for NYUDv2 is avaliable at http://wangkaiwei.org/file/NYUDv2.zip .

The pre-processed NYUDv2 dataset with .npy format is avaliable at https://pan.baidu.com/s/1nxys4pdT4pacWLMScIGT4w with password：k6rv

python ACNet_eval_nyuv2.py --cuda --last-ckpt '.\model\ckpt_epoch_1195.00.pth'

Note: The code is partially based on RedNet (https://github.com/JindongJiang/RedNet) and (https://github.com/warmspringwinds/pytorch-segmentation-detection)