Sense: A Near-Duplicate NFT Detection Protocol on the Pastel Network.
Sense is a lightweight protocol on the Pastel Network, built to assess the relative rareness of a given NFT against near-duplicate meta-data. Sense can recognize even the most subtle similarities between two digital collectibles, even if one has been transformed. The protocol goes beyond the standard “digital fingerprint” approach to establishing the rareness of an NFT, and actually looks at the rareness of the pixel patterns in data. While digital fingerprints do allow users to verify that an NFT was created by a particular creator, this is a fairly weak form of rareness. Sense solves this problem by assigning a ‘Relative Rareness Score’ to quantify how rare an NFT is relative to all NFTs in the underlying dataset.
This score is a number between 0% (i.e., the NFT is identical to an existing NFT) to 100% (i.e., the NFT is not even similar to any known NFT). There are two properties of Sense’s rareness scores that make it far more powerful and useful than existing techniques:
- It does not require an NFT to be an exact duplicate; in fact, the NFT data can be transformed in all sorts of complex ways and still be detected as being a “near-duplicate” of an existing NFT. For instance, the data could be cropped, rotated, stretched, flipped, have colors changes, have random noise added, be inverted, etc., and the system will still “see through” these superficial changes.
- It allows for much more gradation of rareness; rather than reducing the question to a binary “yes/no,” the output provides for a well-defined rareness score to the nearest hundredth of a percent, providing useful information about how visually similar each NFT is compared to all known NFTs on the blockchain.
Example Output of the Sense Protocol