Stereo and Optical flow

Installation Guide

• Install dependencies: numpy, opencv2, MXNet.
• If the MXNet version which you used doesn't support correlation1D operator, you should install MXNet with additional operatior -- edit EXTRA_OPERATORS in config.mk to include the flow_stereo/operator folder.
• Run make all to build extension
• Edit all fields in data/config.py, otherwise it will not work.

Project Organization

• data.dataloader: Data Loader
• data.dataset: Dataset class
• data.augmentation: Pipeline for augmentation
• predict.pipeline: Pipeline for prediction
• docs: Dataset documents
• symbol: Network symbols and customed operators
• others: visualization, metric, and other utilities

Training

Run python -m flow_stereo.train exp_name EXPERIMENT_NAME --gpus GPU_INDEX --epoch RESUMING_EPOCH

Testing Example

Please refer to predict/predict_kitti.py and predict/predict_video.py. Those examples provides a step-by-step walkthrough to help you learn the usage of prediction pipeline