As a seasoned crypto investor and avid follower of the Ethereum ecosystem, I find Vitalik Buterin’s recent endorsement of TiTok AI an intriguing development. The potential of this new image compression method to enable easier creation of profile pictures and NFTs is an exciting prospect for the community.
In the words of Ethereum co-founder Vitalik Buterin, the innovative image compression technique called TiTok AI by Token for Image could compress images to a manageable size, enabling their addition to blockchain networks like Ethereum (ETH).
Buterin referred to the image compression method as a novel means of “encoding a profile picture” on his Warpcast social media platform. If this method can reduce an image to just 320 bits, which he described as “almost equivalent to a hash,” it will make the pictures small enough for inclusion on the blockchain for each user.
As a researcher in the field of blockchain technology and artificial intelligence, I came across an intriguing X-post on Leonardo AI, a renowned image generator platform. The Ethereum co-founder’s attention was piqued by this post, which showcased TikTok’s advanced AI capabilities.
Ethan\_smith\_20, a researcher using that handle, shared concisely how this technique can aid individuals seeking to represent intricate visual information containing high-frequency details into just 32 symbols or tokens.
According to Buterin’s viewpoint, this approach would simplify the process for both developers and creators in generating profile pictures and non-fungible tokens (NFTs).
Solving previous image tokenization issues
The AI system named TiTok, a collaboration between ByteDance, TikTok’s parent company, and the University of Munich, stands out with its unique, one-dimensional tokenization approach. This is different from the commonly used two-dimensional methods in the field.
Based on a study on image tokenization in TikTok, artificial intelligence reduces 256×256-pixel rendered images into just 32 unique representations or tokens.
The paper highlighted shortcomings in past image tokenization techniques, including VQGAN. Prior to this, it was feasible to tokenize images, but methods were constrained by the use of “2D latent grids with predetermined downsampling rates.”
In simpler terms, 2D image segmentation struggled with dealing to redundancies within pictures, and adjacent areas often appeared quite similar.
TikTok employs AI technology to address image issues by breaking down images into 1D sequences of latent representations, which compactly represent the essential information and eliminate redundant regions.
Additionally, implementing a tokenization strategy for images may simplify image management on blockchain systems and significantly boost processing efficiency.
As a crypto investor, I’m excited about this new technology that boasts speeds up to 410 times faster than what we currently have. This is a significant leap forward in computational efficiency for me and the crypto community as a whole.
Read More
Sorry. No data so far.
2024-06-16 18:06