Seam Erasure

Seamlessly erase seams from your favorite 3D models.

Textures seams often produce errors when bi-linearly interpolated. This results in a visible seam line or other undesired artifacts. The goal of this project is to devise a numerical solution to this problem by minimizing the difference between interpolated values of edge pairs. Works for various texture maps regardless of useage (e.g. color maps, normal maps, displacement maps, ambient occlusion, geometry images, and skinning weight textures).

Read more:

• Research Project Page
• Paper
• Video
• Seam Aware Decimation Code

Installation

To install Seam Erasure use the following command in your terminal:

pip install seam-erasure

This is the preferred method to install Seam Erasure.

If you perfer to use Seam Erasure without installing it, you can clone the repo:

git clone https://github.com/zfergus/seam-erasure.git

Dependencies

If you install Seam Erasure using pip all dependencies will be installed. If you choose to use Seam Erasure without installing, you can install all required dependencies using pip install -r requirements.txt.

Dependencies include:

• numpy: linear algebra
• scipy: sparse matrix operations
• pillow: saving/loading texture image files
• recordclass: simple data objects
• tqdm: fancy progressbars

Usage

If you install using pip a seam-erasure tool will be installed. You can use Seam Erase from the command line using the following command:

seam-erasure path/to/input_model path/to/input_texture [-h] [-o path/to/output_texture] [-g] [--sv {none,texture,lerp}] [-d]

Positional arguments:

• path/to/input_model: Path to input mesh file.
• path/to/input_texture: Path to input texture image or directory to load all textures from.

Optional arguments:

• -o path/to/output_texture, --output path/to/output_texture: Name of output texture or directory to save batch textures.
• -g, --global: Should the minimization have global effects? (default: False)
  • This should be used if the texture has global discontinuities. This will propagate changes from the seam inward (see the teaser image for a example of global discontinties).
• --sv {none,texture,lerp}: What method should be used to compute the seam value energy? (default: none)
  • none: do not use a seam value term
  • texture: use difference in original texture
  • lerp: use linearly interpolated values along the edge
    • Values are provided at the vertices in the .obj as additional entries after the xyz triplet (e.g. v <x> <y> <z> [<r>] [<g>] [<b>] [<a>] ... where the additional channels are optional and must match the number of channels in the texture image).
• -d, --data: Should the input texture(s) be loaded as a .data files? (default: False)

Note: if you did not install the code replace seam-erasure in the above command with python main.py.

Files

• seam_erasure/: Python package for Seam-Erasure
• main.py: Command-line interface for seam erasure.
• server.py: Flask based Python code for handling web inputs.
• examples/: examples 3D model and texture files for testing
• static/: Static web page content including style sheets
• templates/: HTML template pages

Web Browser UI

This repository also includes a Flask based server implementation that can be run locally. This provides a simple webpage interface to provide model/texture input and select options. To get this user interface run:

python server.py

This will start a server on the localhost. Navigate to the outputted address in your choice of web browser to view the interface.

Examples

Diffuse Textures

Before	After

Normal Maps

Before	After

Ambient Occlusion

Before	After

Geometry Images

Before	After