-2020.09.22
added dependency
fps option added.
use argument like this
python main.py --fps 30
My Dataset is 2.33Tb. Using OpenCV is too slow. So I use ffmpeg instead.
-
for windows
download ffmpeg build for windows https://www.gyan.dev/ffmpeg/builds/packages/ and add binary exe file's directory to windows environment variable path
-
for linux
sudo apt-get install ffmpeg
-2020.09.19
This is simple frame extractor
I use it for our lab's dataset but this is not public yet.
So I tested on one of the famouse and light benchmark dataset for human action recognition (HMDB-51 dataset).
https://serre-lab.clps.brown.edu/resource/hmdb-a-large-human-motion-database/#Downloads
And it worked well.
FPS option will be added soon
If you follow the details below, the extractor will work on any dataset.
input format :
The dataset must be in one folder.
output foramt :
output folder will be generated like this.
destination_dir/video_name/frame000001.jpg
if you want to change this format please modify line below
if not(os.path.isdir(dst+"\\"+label+"_"+name)):
os.makedirs(os.path.join(dst+"\\"+label+"_"+name))
os.system("ffmpeg -i "+vid+" "+"-r "+str(fps)+" "+ "-start_number 0 "+dst+"\\"+label+"_"+name+"\\frame%6d.jpg")
