paper link : Denoising Diffusion Probabilistic Models
Diffusion models are a class of generative models that excel in image generation by transforming random noise into detailed images through a process inspired by physical diffusion. This involves a forward process where images are gradually noised, and a reverse process where a neural network learns to denoise, effectively generating images from noise.
Key points about diffusion models include:
- High-Quality Images: They produce high-resolution, realistic images.
- Versatile Applications: Useful for tasks like image synthesis, super-resolution, inpainting, and conditional generation.
- Stable Training: More stable and less prone to issues like mode collapse compared to other generative models, such as GANs.
- Computational Intensity: The iterative denoising process requires significant computational resources.
Despite their computational demands, diffusion models' ability to generate diverse and high-quality images makes them a powerful tool in the field of image generation.
- create conda enviroment
$ conda create --name myenv python=3.9 -y - clone the project
$ git clone [link]
- install requirements please install pytorch from
#activate the conda env
$ conda activate [env_name]
# cd into root of the project
$ cd [root of the projcet ]
#then run
$ pip install .
# install the requirements
$ pip install -r requirments.txt
just changer the dataset dir path in the config.yaml file
$ python train_pipeline.py
https://drive.google.com/drive/folders/11Ej5R4HSvo1naScvmAoTaf80FgpEl36d?usp=sharing download pre-train model weight from this like and replace model_ckpt = "{weight_dir}"
$ python predicton_pipe;ine.py
Instructions on how to use the project can be found here.
Guidelines on how to contribute to the project can be found here.
Information about the project's license can be found here.