iriya-ufo/prednet

PredNet

This is a implimentation of PredNet on chainer.

#Testd on *Ubuntu 14.04
*Python 2.7.6
*chainer 1.9.1
*CUDA Toolkit 7.5
*cuDNN v5

#Demo

1. git clone
   $ git clone https://github.com/quadjr/PredNet.git
   $ cd PredNet

2. Prepare Dataset
   This command will download The KITTI Dataset(about 47GB), unzip, and make image lists.
   $ ./scripts/prepare_kitti.sh
   You can intterupt and resume downloading.

3. Train
   On a CPU
   $ python main.py -i dataset/train_list.txt
   On a GPU
   $ python main.py -i dataset/train_list.txt -g 0

Model files and some images will be generated in models/ and images/ directory.
Image suffix means x (input), y (predicted), z (correct).

4. Test
   $ python main.py -i dataset/test_list.txt --test --initmodel models/???.model
   Plase specify the model file with option --initialmodel.
   Test result will be generated in images/ directory.
   Image suffix means x (input), y (predicted).

#TODO *Confirm the learning strategy.