Skip to content

Latest commit

 

History

History
95 lines (76 loc) · 2.83 KB

PREPARE_DATA.md

File metadata and controls

95 lines (76 loc) · 2.83 KB
Raw

Download dataset

DAVIS2016

To download at DAVIS.

wget https://graphics.ethz.ch/Downloads/Data/Davis/DAVIS-data.zip
unzip DAVIS-data.zip

The dataset should be organized as

./data/DAVIS/├── JPEGImages/
                ├── 1080p/(50 category folders)
                ├── 480p/(50 category folders)
             ├── Annotations/
                ├── 1080p/(50 category folders)
                ├── 480p/(50 category folders)
             ├── FlowImages/
                ├── 1080p/(50 category folders)
                ├── 480p/(50 category folders)
             ├── train_vid.npy
             ├── val_vid.npy
                        ...

FBMS

To download at FBMS

wget https://lmb.informatik.uni-freiburg.de/resources/datasets/fbms/FBMS_Trainingset.zip
wget https://lmb.informatik.uni-freiburg.de/resources/datasets/fbms/FBMS_Testset.zip
python ./data/FBMS/FBMS_clean.py

The dataset should be organized as

./data/FBMS/ ├── JPEGImages/
             ├── Annotations/
             ├── train_vid.npy
             ├── val_vid.npy
             ├── RAFT_FlowImages_gap3
             ├── ARFlow_FlowImages_gap3 (optinal)
                        ...

SegTrackv2

To download at SegTrackv2

wget https://web.engr.oregonstate.edu/~lif/SegTrack2/SegTrackv2.zip
python ./data/SegTrackv2/SegTrack_clean.py

The dataset should be organized as

./data/SegTrackv2/├── GroundTruth/
                  ├── ImageSets/
                  ├── train_vid.npy
                  ├── val_vid.npy
                  ├── RAFT_FlowImages_gap1
                  ├── ARFlow_FlowImages_gap1 (optinal)
                        ...

Generate Optical Flow using RAFT

Taken from: https://github.com/charigyang/motiongrouping/tree/main/raft

Please modify the datapath in run_inference.py and generate optical flow for each dataset separately. Note that using gap=3 forFBMS dataset.

cd raft
python run_inference.py

Generate Optical Flow using ARFlow

Download ARFlow from: https://github.com/lliuz/ARFlow.

Please follow the instruction in ARFlow, to generate the environment. Their code has been developed under Python3, PyTorch 1.1.0 and CUDA 9.0 on Ubuntu 16.04.

# Install python packages
pip3 install -r requirements.txt

Please replace the original inference.py, modify the datapath in run_inference.py and generate optical flow for each dataset separately. Note that using gap=3 forFBMS dataset.

git clone https://github.com/lliuz/ARFlow.git
cd ARFlow
python run_inference.py