/GENERATIVE-IMAGE-COMPARISION

Different Evaluation Metrics for Image Generation Models

Primary LanguageJupyter NotebookMIT LicenseMIT

GENERATIVE-IMAGE-COMPARISION Models

This framework utilizes various evaluation metrics to provide a combined and normalized score that assesses the quality and relevance of generated images based on input prompts.

Project Overview

The framework evaluates images using the following metrics:

  • Aesthetic Score: Evaluates the visual appeal of the image.
  • CLIP: Measures the similarity between the image and the input text prompt.
  • QAlign: Provides quality and aesthetics alignment scores for the image.
  • SAM CLIP: Segments the image and evaluates the relevance of segmented parts to the prompt.
  • IQA PyTorch: (Optional) Provides an Image Quality Assessment if weights are available.

Quality_Assurance_framework

Components

  • Input Image: The image to be evaluated.
  • Input Prompt: The text description to which the image is compared.
  • Model Weights: Pre-trained weights for the models used in the evaluation.

Evaluation Process

  1. Input Image and Prompt: Provide the image and the corresponding text prompt.
  2. Evaluation by Models: Various models evaluate the image based on different metrics.
  3. Aesthetic Score: Calculates the aesthetic appeal of the image.
  4. CLIP: Computes the similarity between the image and the text prompt.
  5. QAlign: Provides quality alignment scores.
  6. SAM CLIP: Segments the image and evaluates the relevance of each segment.
  7. IQA PyTorch: (Optional) Provides an Image Quality Assessment if weights are available.
  8. Combined and Normalized Score: The scores from all evaluations are combined and normalized to provide a final score between 0 and 1.

Setup and Installation

  1. Clone the Repository:

    git clone https://github.com/AYUSH27112021/GENERATIVE-IMAGE-COMPARISION.git
cd "GENERATIVE-IMAGE-COMPARISION"

  2. Install Dependencies: Ensure you have Python installed. Install the required Python packages:

    pip install -r requirements.txt

  3. Download Model Weights: Download and place the pre-trained model weights in the appropriate directory as needed.

Usage

Command Line Interface

To compute the similarity between an image and a text prompt using all metrics at once:

python All_metrics.py <image_path> <text_input>

Example

python compute_similarity.py "assets\areoplane_prompt.jpg" "A picture of an airplane in a thunderstorm.

NOTE:

  1. To run the ALL_metric.py file, the GPU must support CUDA, and PyTorch must be compiled with FlashAttention.
  2. The ALL_metric.py file requires a sufficient amount of GPU memory to load and run the models.
  3. Some models also support torch.run and can be executed in parallel with others.
  4. Models can also be run separately. See the metric folder README for more details.