spatiotemporal convolutions for video analysis and study their effects on action recognition using the pre trained resnet model
The directory tree strucuture is :
folder/
-------------action_classify.ipynb
-------------dataset/
-------------models/
-------------data_imgset/
-------------myutils.py
the dataset has beem downloaded from which is available at https://serre-lab.clpsbrown.edu/resource/hmdb-a-large-human-motion-database/#overview.
The HMDB dataset was collected from various sources, including movies, the Prelingerarchive, YouTube, and Google videos. It is a pretty large dataset (2 GB) with a total of 7,000 video clips. There are 51 action classes, each containing a minimum of 101 clips.
extract the hmdb51_org.rar file into the dataset folder. The folder should contain 51 subfolders corresponding to 51 class actions. Also, each subfolder should contain at least 101 video files of the .avi type for each action class.