PatchMatch algorithm

Implementation of the PatchMatch algorithm proposed by Barnes, Shechtman, Finkelstein and Goldman. SOURCE: http://gfx.cs.princeton.edu/gfx/pubs/Barnes_2009_PAR/patchmatch.pdf

Prerequisites

Software

• OpenCV [http://opencv.org/]
• CMake [http://www.cmake.org/install/]

Datasets

• Middlebury dataset [http://vision.middlebury.edu/flow/data/comp/zip/other-color-twoframes.zip ]
• Ground-truth dataset [http://vision.middlebury.edu/flow/data/comp/zip/other-gt-flow.zip]
• Evaluation tool [http://vision.middlebury.edu/flow/code/flow-code.zip]

Build instructions

PatchMatch algorithm

• git-clone the repo:

git clone https://github.com/alexanderb14/PatchMatch-Optical-Flow.git

• cd into it:

cd PatchMatch-Optical-Flow

• create a build directory and cd into it:

mkdir build && cd build

• run cmake and initiate the build process:

cmake .. && make

Evaluation tool

• build imageLib

cd imageLib make

• build evaluation tool

cd .. make

Usage

• two executables will be generated
  • PatchMatchApp
    • gui program, intended to test similarity measures
    • user can click to a location in each picture. Then, the similarity is calculated.
  • PatchMatchBenchmark
    • headless program, intended to work with gen_dataset.sh:
      • If you configure gen_dataset.sh in the right way, it will generate you all flow images (PatchMatch+Ground-truth) for the whole dataset, using two similarity measures. To do this, modify the following 3 variables in the header of the file:

        • FLOWCODE_EXEC=

        • DATASET_DIR=<directory of middlebury dataset (where folders like Dimetrodon, Grove2, ... are located)>

        • GROUNDTRUTH_DIR=<directory of ground-truth (where folders like Dimetrodon, Grove2, ... are located)>

---

---

Here are the results of PatchMatch algorithm implementation on the middlebury dataset. For the first images CrossCorrelationNormalized, for the second images SumOfSquaredDifferences have been utilized.

---

---

---

---

---

---

---

---

---

---

---

---