Deep Predictive Coding Network (PredNet) implemented with Chainer v4
You must use a GPU to train.
- Python >=3.6
- Chainer 4.0
- opencv-python
- matplotlib
- tqdm
- glob
- shutil
Download preprocessed KITTI datasets from prednet_kitti_data.zip.
You have to convert hickle-python2 data to npy data.
When you use GPU, please run following command.
python train.py -g 0
When you want to train extrapolation model, run
python train_extrap.py -g 0
Run following command.
python generate_result_images.py -g 0
or
python generate_result_images_extrap.py -g 0
- Deep predictive coding networks for video prediction and unsupervised learning
- Unsupervised Learning of Visual Structure using Predictive Generative Networks
- Long short-term memory
- Convolutional LSTM network: a machine learning approach for precipitation nowcasting
- Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects