Home

Awesome

Few-shot Stereo Matching with High Domain Adaptability Based on Adaptive Recursive Network

New: Our testing code is currently under development and will be released soon.

This repository contains the code for the paper "Few-shot Stereo Matching with High Domain Adaptability Based on Adaptive Recursive Network". The training code is now released. Our testing code will be publicly available soon.

Getting Started and Example Usage

Step 1: Prepare the Training Data

To begin, run the get_training_data.py script. This script will randomly sample image pairs from the SceneFlow directory and create the training dataset. The images will be saved in the training_patchs folder.

Note: Use --root_dir /path/to/your/SceneFlow/ to specify your SceneFlow directory.

Run the script:

python get_training_data.py --root_dir /path/to/your/SceneFlow/ --num_processes 4 --images_num 500

Step 2: Prepare the Testing Data

Similarly, run the get_testing_data.py script to sample image pairs for the testing dataset. The images will be saved in the testing_patch folder.

Note: Use --root_dir /path/to/your/SceneFlow/ to specify your SceneFlow directory.

Run the script:

python get_testing_data.py --root_dir /path/to/your/SceneFlow/ --num_processes 4 --images_num 500

Step 3: Train the Model

After generating the training and testing datasets, you can start training the model by running train.py. This script will read the training and testing data from the training_patchs and testing_patch folders, respectively. The trained model will be saved in the good_model directory.

Run the training script:

python train.py --epoch 500

Training Script Command-line Arguments

The train.py script supports the following command-line argument:

Directory Structure