The five video classification methods:
- Classify one frame at a time with a ConvNet
- Extract features from each frame with a ConvNet, passing the sequence to an RNN, in a separate network
- Use a time-dstirbuted ConvNet, passing the features to an RNN, much like #2 but all in one network (this is the
lrcnnetwork in the code). - Extract features from each frame with a ConvNet and pass the sequence to an MLP
- Use a 3D convolutional network (has two versions of 3d conv to choose from)
See the accompanying blog post for full details: https://medium.com/@harvitronix/five-video-classification-methods-implemented-in-keras-and-tensorflow-99cad29cc0b5
This code requires you have Keras 2 and TensorFlow 1 or greater installed. Please see the requirements.txt file. To ensure you're up to date, run:
pip install -r requirements.txt
You must also have ffmpeg installed in order to extract the video files. If ffmpeg isn't in your system path (ie. which ffmpeg doesn't return its path, or you're on an OS other than *nix), you'll need to update the path to ffmpeg in data/2_extract_files.py.
First, download the dataset from UCF into the data folder:
cd data && wget http://crcv.ucf.edu/data/UCF101/UCF101.rar
Then extract it with unrar e UCF101.rar.
Next, create folders (still in the data folder) with mkdir train && mkdir test && mkdir sequences && mkdir checkpoints.
Now you can run the scripts in the data folder to move the videos to the appropriate place, extract their frames and make the CSV file the rest of the code references. You need to run these in order. Example:
python 1_move_files.py
python 2_extract_files.py
Before you can run the lstm and mlp, you need to extract features from the images with the CNN. This is done by running extract_features.py. On my Dell with a GeFore 960m GPU, this takes about 8 hours. If you want to limit to just the first N classes, you can set that option in the file.
The CNN-only method (method #1 in the blog post) is run from train_cnn.py.
The rest of the models are run from train.py. There are configuration options you can set in that file to choose which model you want to run.
The models are all defined in models.py. Reference that file to see which models you are able to run in train.py.
Training logs are saved to CSV and also to TensorBoard files. To see progress while training, run tensorboard --logdir=data/logs from the project root folder.
I have not yet implemented a demo where you can pass a video file to a model and get a prediction. Pull requests are welcome if you'd like to help out!
- Add data augmentation to fight overfitting
- Support multiple workers in the data generator for faster training
- Add a demo script
- Support other datasets
- Implement optical flow
- Implement more complex network architectures, like optical flow/CNN fusion
Khurram Soomro, Amir Roshan Zamir and Mubarak Shah, UCF101: A Dataset of 101 Human Action Classes From Videos in The Wild., CRCV-TR-12-01, November, 2012.