Continuous Cost Aggregation (CCA) for Dual-Pixel Disparity Extraction

Link to paper: https://arxiv.org/abs/2306.07921

Description folders

1. CSGM : folder containing all implementations divided based on data used (DP-DSLR/PHONE, Stereo).
2. demo_DP: folder with different demo files for DP data: DSLR + Google phone
3. demo_Stereo: folder with demo for traditional stereo - comparison between C-SGM and SGM. Filter is the same for both SGM and C-SGM (names are changed).
4. DP_data_example: example data of DP-DSLR/PHONE
5. Stereo_data_example: example data of Stereo middlebury at Quarter resultion.
6. evaluate_funcs: functions used for evaluation
7. openCvMatlab : this is complied for windows, might need to mex-install in different eviorment

Demos:

1. demo_csgm_DSLR_quantitive - test on DSLR data set (1 in data)
2. demo_csgm_phone+filter_test_phone.py: test on phone data (2 in data).
3. demo_csgm_stereo+bilaterl_filter_csgm.py: test on middlebury data (3 in data).
4. demo_sgm_stereo+bilaterl_filter_sgm.py: Adapted SGM on middlebury data (3 in data).

How to use

1. Add all files to path.
2. Run one of the above demos.
3. OpenCV for Matlab might need to be installed to load images.

Data:

1. DP-data: There is an example for one image from ICCP-2020 paper (DSLR): DP_data_example\ICCP2020\GT_data. Rest of data can be found in: https://github.com/abhijithpunnappurath/dual-pixel-defocus-disparity/blob/master/README.md

2. Google-phone data: There is one example from google data from ICCV 2019 paper: DP_data_example\google2019\test Rest of data can be found in: https://github.com/google-research/google-research/blob/master/dual_pixels/README.md

3. Stereo middlebury: https://vision.middlebury.edu/stereo/submit3/

Different algorithms (compared in paper):

1. DLP: Learning Single Camera Depth Estimation using Dual-Pixels. Implementation not available, predictions on google data set found in: https://github.com/google-research/google-research/blob/master/dual_pixels/README.md

2. SDoF: Synthetic Depth-of-Field with a Single-Camera Mobile Phone Implementation not available.

3. DPE: Dual Pixel Exploration: Simultaneous Depth Estimation and Image Restoration Implementation in: https://github.com/panpanfei/Dual-Pixel-Exploration-Simultaneous-Depth-Estimation-and-Image-Restoration

4. DPdisp: Modeling Defocus-Disparity in Dual-Pixel Sensors Implementation found in: https://github.com/abhijithpunnappurath/dual-pixel-defocus-disparity/blob/master/README.md

5. SGM: SGM implementation, customed to use BT score instead of SAD. Implementation adapted from: https://github.com/kobybibas/semi_global_matching