Skip to content

Latest commit

 

History

History
140 lines (119 loc) · 4.39 KB

installation.md

File metadata and controls

140 lines (119 loc) · 4.39 KB
Raw

Installation

The project is based on the pytorch 1.8.1 with python 3.8.

1. Clone the Git repo

$ git clone https://github.com/tianyu0207/PEBAL.git
$ cd pebal

2. Install dependencies

  1. create conda env 
    $ conda env create -f pebal.yml
  2. install the torch 1.8.1 
    $ conda activate pebal
# IF cuda version < 11.0
$ pip install torch==1.8.1+cu102 torchvision==0.9.1+cu102 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html
# IF cuda version >= 11.0 (e.g., 30x or above)
$ pip install torch==1.8.1+cu111 torchvision==0.9.1+cu111 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html

3. Prepare dataset

cityscapes

  1. please download the Cityscapes dataset (gt_Fine).
  2. (optional) you might need to preprocess Cityscapes dataset in here, as we follow the common setting with 19 classes.
  3. specify the COCO dataset path in code/config/config.py file, which is C.city_root_path.

fishyscapes

  1. for the time being, you can download from the official website in here.
  2. specify the coco dataset path in code/config/config.py file, which is C.fishy_root_path.

You can alternatively download both preprocessed fishyscapes & cityscapes datasets here (token from synboost GitHub).

coco (for outlier exposures)

  1. please follow Meta-OoD to prepare the COCO images serving as OoD proxy for OoD training. This script generates binary segmentation masks for COCO images not containing any instances that could also be assigned to one of the Cityscapes (train-)classes. Execute via: 
    $ python preparation/prepare_coco_segmentation.py
  2. specify the coco dataset path in code/config/config.py file, which is C.coco_root_path.

4. Dataset Structure

  1. the tree structures of the training datasets (including both cityscapes, and coco) are shown below.
city_scape/
├── annotation
│   └── city_gt_fine
│       ├── train
│       └── val
└── images
    └── city_gt_fine
        ├── train
        └── val
coco/
├── annotations
│   └── ood_seg_train2017
└── train2017
  1. the tree sturcutres of all the validation datasets are shown in below.
fishyscapes/
├── LostAndFound
│   ├── entropy
│   ├── labels
│   ├── labels_with_ROI
│   ├── logit_distance
│   ├── mae_features
│   ├── original
│   ├── semantic
│   └── synthesis
└── Static
    ├── entropy
    ├── labels
    ├── labels_with_ROI
    ├── logit_distance
    ├── mae_features
    ├── original
    └── semantic

road_anomaly/
├── entropy
├── labels
├── mae_features
├── original
├── semantic
└── synthesis

lost_and_found/
├── gtCoarse
│   ├── test
│   │   ├── 02_Hanns_Klemm_Str_44
│   │   ├── 04_Maurener_Weg_8
│   │   ├── 05_Schafgasse_1
│   │   ├── 07_Festplatz_Flugfeld
│   │   └── 15_Rechbergstr_Deckenpfronn
│   └── train
│       ├── 01_Hanns_Klemm_Str_45
│       ├── 03_Hanns_Klemm_Str_19
│       ├── 06_Galgenbergstr_40
│       ├── 10_Schlossberg_9
│       ├── 11_Parkplatz_Flugfeld
│       ├── 12_Umberto_Nobile_Str
│       ├── 13_Elly_Beinhorn_Str
│       └── 14_Otto_Lilienthal_Str_24
└── leftImg8bit
    ├── test
    │   ├── 02_Hanns_Klemm_Str_44
    │   ├── 04_Maurener_Weg_8
    │   ├── 05_Schafgasse_1
    │   ├── 07_Festplatz_Flugfeld
    │   └── 15_Rechbergstr_Deckenpfronn
    └── train
        ├── 01_Hanns_Klemm_Str_45
        ├── 03_Hanns_Klemm_Str_19
        ├── 06_Galgenbergstr_40
        ├── 10_Schlossberg_9
        ├── 11_Parkplatz_Flugfeld
        ├── 12_Umberto_Nobile_Str
        ├── 13_Elly_Beinhorn_Str
        └── 14_Otto_Lilienthal_Str_24