watermark-removal

Why we included this project

Watermarks baked into an image are hard to remove cleanly; simple crops or blurs leave obvious damage. This project treats the job as an inpainting problem, using a deep learning model to fill in covered pixels by borrowing texture and structure from the surrounding image, with ideas drawn from Contextual Attention (CVPR 2018) and Gated Convolution (ICCV 2019). The removed watermark is hard to tell apart from the original. The repo includes a command-line entry point and a Docker image, but you supply the pre-trained TensorFlow checkpoint yourself, and the code targets TensorFlow 1.15, so it reads as a solid starting point rather than a turnkey service. Developers prototyping inpainting pipelines and researchers tracing how these restoration models behave will get the most from the small, readable codebase.