The paper "Structured Output Learning with Conditional Generative Flows" by You Lu and Bert Huang. proposes a new approach for structured output learning, called conditional Glow (c-Glow), which is a conditional generative flow model that can directly model the conditional distribution p(y|x) without restrictive assumptions. The authors show that c-Glow can be trained efficiently by exploiting the fact that invertible flows allow exact computation of log-likelihood, and that it can generate structured outputs that are comparable in quality to those produced by state-of-the-art deep structured prediction approaches. The paper evaluates c-Glow on five structured prediction tasks: binary segmentation, multi-class segmentation, color image denoising, depth refinement, and image inpainting. The results demonstrate the effectiveness of c-Glow in handling a variety of structured prediction problems.
To dive into the transformative world of cGLOW, begin by setting up your environment with these steps. We recommend using a virtual environment for an isolated setup.
-
Clone the repository
git clone https://github.com/Manuelnkegoum-8/cGLOW.git cd cGLOW -
Set up a virtual environment (optional but recommended)
- For Unix/Linux or MacOS:
python3 -m venv env source env/bin/activate - For Windows:
python -m venv env .\env\Scripts\activate
- For Unix/Linux or MacOS:
-
Install the requirements
pip install -r requirements.txt
-
Usage
-
To train the cGLOW model with default settings:
./train.sh
-
For generating images using a pre-trained model:
./inference.sh
-
This code is used for the experiments of binary segmentation on the Retina Blood Vessel dataset
