Release Notes

• Code refactored from the official repository from DAI-Lab for ease of use and simplicity
• Watermark encoding and decoding process have gained speed uplift (3.5~4x faster)
• Incorporated model weights which was trained in 32-bit data
• Revert make_pair function to its original code due to instability issue during training
• Users are now able to create their own data watermark instead random generated data for encoding