Title: "Reducing Mosaic Artifacts in Deep Super-Resolution Networks"
Authors: Jianfeng Zhang, Liwei Wang, Yuchen Fan (example) — note: if authors differ, search the exact title.
Why it’s significant: This paper presents practical methods to reduce mosaic (blocky) artifacts that commonly appear when applying super-resolution or denoising models to compressed or mosaiced inputs. It combines perceptual loss, frequency-domain regularization, and a training curriculum that prioritizes edge preservation, yielding visually coherent outputs without oversmoothing.
Key contributions (useful takeaways):
Before deep learning, users spent hours on:
If you “spent your S” (time, sanity, software subscriptions) on these, you know their painful limitations: You cannot recover lost information; you can only hide blocks. ds ssni987rm reducing mosaic i spent my s updated
You’ve “reduced mosaic” significantly, especially for compression artifacts. For intentional mosaic (SSNI... type content), you will need inpainting models like LaMa or MAT, but they are less reliable.
For this example, we use RealESRGAN (best for compression mosaics). If you “spent your S” (time, sanity, software
git clone https://github.com/xinntao/Real-ESRGAN.git
cd Real-ESRGAN
python inference_realesrgan.py -i frames/ -o output_frames/ -s 2 --outscale 2
The field is moving fast. Expect:
If you “spent your S” on old methods, update now to RealESRGAN-animevideo or BasicVSR++ for video. or older MPEG-2
When video is compressed with codecs like H.264, H.265, or older MPEG-2, the image is divided into 8×8 or 16×16 pixel blocks. At low bitrates, these blocks become visible as a “mosaic” — square patterns, especially in flat or dark areas.
Example: A blue sky turns into a checkerboard of slightly different blue blocks.