search

HHTseng / video-classification

Tutorial for video classification/ action recognition using 3D CNN/ CNN+RNN on UCF101
916 stars 216 forks source link

data as img or video #17

Open nicholasguimaraes opened 4 years ago

nicholasguimaraes commented 4 years ago

Hello, I tried downloading the dataset from the link you provided in your repository. I couldn't extract it but I realized that the data divided into .jpg frames.

I downloaded the UCF-101 dataset from their own website and the data is in .avi

Is there a problem using the videos instead of images as training data?

mampferd commented 4 years ago

Hi, I can answer how to get the .jpg files via the .001/.002/.003 files. You need to have all the three of them in the same folder and then extract via 7-zip or similar programs the data just from the .001 file. The information will be merged and the .jpg-files extracted automatically.

Hope this can still help :)

baniks commented 4 years ago

@HHTseng Hi, I was getting error while extracting the data file. I came across this post later. Anyways, I ended up downloading the original UCF101 dataset and saving the frames from it. It would be really helpful if you can mention the folder structure of the data directory in the description. I had to play around, till I found out the correct directory structure.

jpegs_256
├── v_ApplyEyeMakeup_g01_c01
│   ├── frame000001.jpg
│   ├── frame000002.jpg
│   ├── frame000003.jpg
│   .
.
.
└── v_Typing_g25_c04
    ├── frame000001.jpg
    ├── frame000002.jpg
    .

Thank you.