Skip to content

Latest commit

 

History

History
148 lines (115 loc) · 3.67 KB

PREPARE_DATA.md

File metadata and controls

148 lines (115 loc) · 3.67 KB
Raw

How to prepare data

Pre-requisite

  1. We assume that you've cloned this repo under $REPO_DIR

    REPO_DIR='/path/to/clone/this/repo'
git clone https://www.github.com/1Konny/hierarchicalvideoprediction $REPO_DIR

----

./$REPO_DIR
|-- assets
|-- docs
|-- image_generator
|-- scripts
`-- structure_generator

  2. Clone the repo for the semantic segmentation model used in our paper.

    SEMSEG_REPO_DIR='/path/to/clone/this/repo'
git clone https://github.com/1Konny/semantic-segmentation.git --single-branch --branch sdcnet $SEMSEG_REPO_DIR

  3. Download pretrained weights for the sementic segmentation model (Cityscapes, Kitti) to $SEMSEG_REPO_DIR/pretrained_weights directory as follows.

    $SEMSEG_REPO_DIR/pretrained_models/cityscapes_best.pth
$SEMSEG_REPO_DIR/pretrained_models/kitti_best.pth

KITTI Dataset

  1. Go to the official website, download zip files containing videos, and extract all of them under $REPO_DIR/datasets_raw/KITTI/images directory as follows.

    ./$REPO_DIR/datasets_raw/KITTI/images
|-- 2011_09_26
|   |-- 2011_09_26_drive_0001_sync
|   |-- ...
|   `-- 2011_09_26_drive_0119_sync
|-- 2011_09_28
|   |-- 2011_09_28_drive_0001_sync
|   |-- ...
|   `-- 2011_09_28_drive_0225_sync
|-- 2011_09_29
|   |-- 2011_09_29_drive_0004_sync
|   |-- ...
|   `-- 2011_09_29_drive_0108_sync
|-- 2011_09_30
|   |-- 2011_09_30_drive_0016_sync
|   |-- ...
|   `-- 2011_09_30_drive_0072_sync
`-- 2011_10_03
    |-- 2011_10_03_drive_0027_sync
    |-- ...
    `-- 2011_10_03_drive_0058_sync

  2. Extract semantic label maps

    cd $SEMSEG_REPO_DIR
bash extract_labels.sh KITTI $REPO_DIR/datasets_raw/KITTI/images $REPO_DIR/datasets_raw/KITTI/semantic_labels

  3. Pre-process images and labels

    cd $REPO_DIR
python datasets_raw/process_kitti.py

  4. Then, the following directories will be saved:

    $REPO_DIR/structure_generator/datasets/KITTI_64
$REPO_DIR/image_generator/datasets/KITTI_vid2vid_90

Cityscapes Dataset

  1. Go to the official website, download leftImg8bit_sequence_trainvaltest.zip, and extract all of them under $REPO_DIR/datasets_raw/Cityscapes/images directory as follows.

    ./$REPO_DIR/datasets_raw/Cityscapes/images
|-- test
|   |-- berlin
|   |-- bielefeld
|   |-- bonn
|   |-- leverkusen
|   |-- mainz
|   `-- munich
|-- train
|   |-- aachen
|   |-- bochum
|   |-- bremen
|   |-- cologne
|   |-- darmstadt
|   |-- dusseldorf
|   |-- erfurt
|   |-- hamburg
|   |-- hanover
|   |-- jena
|   |-- krefeld
|   |-- monchengladbach
|   |-- strasbourg
|   |-- stuttgart
|   |-- tubingen
|   |-- ulm
|   |-- weimar
|   `-- zurich
`-- val
    |-- frankfurt
    |-- lindau
    `-- munster

  2. Extract semantic label maps

    cd $SEMSEG_REPO_DIR
bash extract_labels.sh Cityscapes $REPO_DIR/datasets_raw/Cityscapes/images $REPO_DIR/datasets_raw/Cityscapes/semantic_labels

  3. Pre-process images and labels

    cd $REPO_DIR
python datasets_raw/process_cityscapes.py

  4. Then, the following directories will be saved:

    $REPO_DIR/structure_generator/datasets/Cityscapes_256x512
$REPO_DIR/image_generator/datasets/Cityscapes_256x512