Not to be confused with social media platform TikTok, this new TiTok compression method significantly reduces image size, making it more practical for blockchain storage.
Buterin was emphasize TiTok’s blockchain potential on decentralized social network platform Farcaster, claiming that “320 bits is essentially a hash function. Small enough to be included in the chain for each user.”
This development could have significant impacts on the digital image storage of profile pictures (PFP) and non-fungible tokens (NFTs).
Compress TiTok images
Developed by researchers from ByteDance and the Technical University of Munich, TiTok allows compressing an image into 32 small pieces of data (bits) without losing quality.
According to article TiTok research, advanced AI image compression allows TiTok to compress a 256×256 pixel image into “32 separate tokens.”
TiTok is a one-dimensional (1D) image encoding framework that “breaks the existing grid constraints in 2D encoding methods,” resulting in cleaner and more flexible images.
“As a result, it leads to significant speedups in the sampling process (e.g., 410 times faster than DiT-XL/2) while still achieving competitive generation quality.”
Machine learning images
TiTok uses advanced machine learning and AI, using transformer-based models to convert images into encoded representations.
This method uses region redundancy, meaning it identifies and uses redundant information in different regions of the image to reduce the overall data size of the final product.
“Recent advances in generative models have emphasized the important role of image encoding in the efficient synthesis of high-resolution images.” According to the research paper, TiTok’s “compact latent representation” can yield “much more efficient and effective representations than conventional techniques.”
TiTok, not TikTok
Despite having a similar name, the social networking platform TikTok did not receive support from Buterin.
The Ethereum co-founder, emphasizing TiTok’s blockchain potential, added credibility to this new AI-driven image compression method.
“Unlike existing 2D VQ models that treat the image latent space as a 2D grid, we provide a cleaner formulation for encoding images into a 1D latent sequence.” The proposed new method can “represent an image with eight to 64 times fewer tokens” than “2D encoders,” and the team hopes this research can shed light on “ more efficient visual representations.”
Thach Sanh
According to Cointelegraph
