From cc93d11a3e3d2fb61a58b7351b245fe0e6164fd5 Mon Sep 17 00:00:00 2001 From: Rishik Mourya Date: Fri, 19 Feb 2021 10:03:36 +0530 Subject: [PATCH] added updates --- README.md | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+) diff --git a/README.md b/README.md index 43f44b4..d548abb 100644 --- a/README.md +++ b/README.md @@ -4,6 +4,33 @@ +## ✅ Update 1 + +Updated the model architecture, and reduced the parameters count from ~3,000,000 to ~500,000 for faster inference. + +## ✅ Update 2 + +Added api endpoint for easy usage: + +```python + +from api import remove_watermark + +remove_watermark( + image_path = IMAGE_NAME, + mask_path = MASK_NAME, + max_dim = MAX_DIM, + show_step = SHOW_STEPS, + reg_noise = REG_NOISE, + input_depth = INPUT_DEPTH, + lr = LR, + training_steps = TRAINING_STEPS, + tqdm_length = 900 +) + +``` + + ## __This is the implementation of paper [Deep Image Prior](https://dmitryulyanov.github.io/deep_image_prior), all credit goes its authors.__ CNNs are very common for image generation and restoration tasks. And it is believed that their great performance is because of their ability to learn realistic image priors from training on large datasets. This paper shows that the structure of a generator alone is sufficient to provide enough low-level image statistics without any learning. Thus most of the image restoration tasks, for example, denoising, super-resolution, artefacts removal, watermark removal etc can be done with highly realistic results without any training.