Denseflow

Extracting dense flow field given a video.

Features

• support multiple optical flow algorithms
• support single video (or a frame folder) / a list of videos (or a list of frame folders) as input
• support multiple output types (image, hdf5)
• faster, 40% faster (by parallelize IO & computation)
• record the progress when extract a list of videos (Note: restart from the recent "done video", that is, the recent "approximately done video" may not actually done)
• 105 fps in RTX2080Ti

Install

Dependencies:

• OpenCV: opencv3 (Install Guide)| opencv4
• CUDA (driver version > 400)
• Boost
• HDF5 (Optional)

git clone https://github.com/sming256/denseflow
cd denseflow
mkdir build && cd build
cmake ..
make -j
sudo make install

Usage

$ denseflow_gpu -h
GPU optical flow extraction.
Usage: denseflow_gpu [params] input
        -a, --algorithm (value:tvl1)
                optical flow algorithm (nv/tvl1/farn/brox)
        -b, --bound (value:32)
                maximum of optical flow
        --cf, --classFolder
                outputDir/class/video/flow.jpg
        -d, --deviceId (value:0)  # may have bug in CUDA10
                set gpu id
        -h, --help (value:true)
                print help message
        --if, --inputFrames
                inputs are frames
        --newHeight, --nh (value:0)
                new height
        --newShort, --ns (value:0)
                short side length
        --newWidth, --nw (value:0)
                new width
        -o, --outputDir (value:.)
                root dir of output
        -s, --step (value:1)
                right - left (0 for img, non-0 for flow)
        -v, --verbose (value:0)
                verbose
        input
                filename of video or folder of frames or a list.txt of those

Extract optical flow of a single video

denseflow_gpu test.avi -b=20 -a=tvl1 -s=1 -v=1

• test.avi: input video
• tmp: folder containing RGB images and optical flow images
• tvl1: optical flow algorithm
• v: verbose
• s: step, extract frames only when step=0

Extract optical flow of a list of videos with single GPU

denseflow_gpu videolist.txt  -b=20 -a=tvl1 -s=1 --nw=224 --nh=224 -v=0 --outputDir=./flow/

• videolist.txt: input videolist.txt
• outputDir: output generated images to folder.

Better to check: resize / class folder / input image

Extract optical flow of video folder with multiple GPUs

cd ./tools
sh extract.sh

• SRC_FOLDER: folder of video source folder
• OUT_FOLDER: flow saving path
• NUM_GPU: number of used GPU, which means each GPU runs a subprocess

Credits

Modified based on yuanjun's fork of dense_flow.

Main Authors:

Shiguang Wang, Zhizhong Li

Modified:

Shuming Liu