The extracted watermark is inaccurate after some processing of the video

[MicoRobot]

I am currently trying to perform secondary editing on videos to test watermark extraction.
I am currently facing a few issues(the following operations are all completed in the "CapCut" application)：

• Resized to 50% of its original size, the extracted watermark is garbled
• Rotated 180 degrees, the extracted watermark is garbled
• Flip the video, the extracted watermark text may have some inaccuracies
• Crop the video edges by 15%, and the remaining 75% of the extracted watermark results are correct. The watermark extracted after trimming the edge by 50% appears garbled

I want to know if these results are correct? I tried different videos using 'inference.av. py' to add and extract watermarks, and the results were similar to the above conclusion.

[hadyelsahar]

Hi @MicoRobot it is normal that the watermark "could" be altered by these operations especially with Capcut also adding upload / download compression. Do you know what is the bit accuracy when the "watermark is garbled" ? If it is still high you can retrieve it using few Error correction code techniques.

[pierrefdz]

• Could you tell us what kind of videos you tested on (notably their resolution), this would help us for the next iterations of the project. We mainly tested on high resolution images, so performance may decrease on low resolution videos.