Angelina8120 / InforMS

Official Implementation of Informative Classes Matter: Towards Unsupervised Domain Adaptive Nighttime Semantic Segmentation

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Informative Classes Matter: Towards Unsupervised Domain Adaptive Nighttime Semantic Segmentation, ACM MM 2023

Setup Environment

The code is run with Python 3.8.13. To install the packages, use:

pip install -r requirements.txt

Set Data Directory

The following environment variable must be set:

export DATA_DIR=/path/to/data/dir

Setup Datasets

Cityscapes

Please, download leftImg8bit_trainvaltest.zip and gt_trainvaltest.zip from here and extract them to $DATA_DIR/Cityscapes.

$DATA_DIR
├── Cityscapes
│   ├── leftImg8bit
│   │   ├── train
│   │   ├── val
│   ├── gtFine
│   │   ├── train
│   │   ├── val
├── ...
ACDC

Please, download rgb_anon_trainvaltest.zip and gt_trainval.zip from here and extract them to $DATA_DIR/ACDC.

$DATA_DIR
├── ACDC
│   ├── rgb_anon_trainvaltest
│   │   ├── rgb_anon
│   │   │   ├── fog
│   │   │   ├── night
│   │   │   ├── rain
│   │   │   ├── snow
│   ├── gt_trainval
│   │   ├── gt
│   │   │   ├── fog
│   │   │   ├── night
│   │   │   ├── rain
│   │   │   ├── snow
├── ...
Dark Zurich

Please, download the Dark_Zurich_train_anon.zip, Dark_Zurich_val_anon.zip, and Dark_Zurich_test_anon_withoutGt.zip from here and extract them to $DATA_DIR/DarkZurich.

$DATA_DIR
├── DarkZurich
│   ├── rgb_anon
│   │   ├── train
│   │   ├── val
│   │   ├── val_ref
│   │   ├── test
│   │   ├── test_ref
│   ├── gt
│   │   ├── val
├── ...
Nighttime Driving

Please, download the NighttimeDrivingTest.zip from here and extract it to $DATA_DIR/NighttimeDrivingTest.

$DATA_DIR
├── NighttimeDrivingTest
│   ├── leftImg8bit
│   │   ├── test
│   ├── gtCoarse_daytime_trainvaltest
│   │   ├── test
├── ...

Training

Coming soon

Testing

Coming soon

Results

To facilitate qualitative segmentation comparisons, validation set and testing set predictions of InforMS can be directly downloaded.

Model Task Test Set Test Score Predictions
InforMS-DeepLabv2 Cityscapes→ACDC-night ACDC-night-test 52.4% mIoU ACDC-night-test
InforMS-DeepLabv2 Cityscapes→ACDC-night ACDC-night-val 44.0% mIoU ACDC-night-val
InforMS-DAFormer Cityscapes→ACDC-night ACDC-night-test 56.9% mIoU ACDC-night-test
InforMS-DAFormer Cityscapes→ACDC-night ACDC-night-val 47.7% mIoU ACDC-night-val
InforMS-HRDA Cityscapes→ACDC-night ACDC-night-test 65.1% mIoU ACDC-night-test
InforMS-HRDA Cityscapes→ACDC-night ACDC-night-val 55.4% mIoU ACDC-night-val
InforMS-DeepLabv2 Cityscapes→Dark Zurich Dark Zurich-test 55.0% mIoU Dark Zurich-test
InforMS-DeepLabv2 Cityscapes→Dark Zurich Dark Zurich-val 40.0% mIoU Dark Zurich-val
InforMS-DeepLabv2 Cityscapes→Dark Zurich Nighttime Driving 53.7% mIoU Nighttime Driving
InforMS-DAFormer Cityscapes→Dark Zurich Dark Zurich-test 57.7% mIoU Dark Zurich-test
InforMS-DAFormer Cityscapes→Dark Zurich Dark Zurich-val 45.1% mIoU Dark Zurich-val
InforMS-DAFormer Cityscapes→Dark Zurich Nighttime Driving 56.0% mIoU Nighttime Driving
InforMS-HRDA Cityscapes→Dark Zurich Dark Zurich-test 64.7% mIoU Dark Zurich-test
InforMS-HRDA Cityscapes→Dark Zurich Dark Zurich-val 52.5% mIoU Dark Zurich-val
InforMS-HRDA Cityscapes→Dark Zurich Nighttime Driving 58.5% mIoU Nighttime Driving

About

Official Implementation of Informative Classes Matter: Towards Unsupervised Domain Adaptive Nighttime Semantic Segmentation


Languages

Language:Python 92.4%Language:C++ 3.9%Language:Cuda 3.7%