Details
Non-fungible tokens have been growing in popularity, but this also means they have become a tempting target for savvy fraudsters. As a result, NFT creators and projects as well as brands and marketplaces have had to deal with a range of violations of their intellectual properties, including:
Attacks via compromised social media accountsTrademark infringementUnauthorized name/art usagePlagiarized workAs these IP violations lead to drastic financial losses and prevent the wider adoption of NFTs, a sophisticated and robust solution is needed to help confront the challenge.
Develop a high-end solution that would allow NFT creators, projects, etc. to spot and prevent the infringement of their intellectual property
Incorporate the ability to identify IP violations by image, trademark, project, brand, or creator name
Enable users to generate comprehensive DMCA reports on authorized IP usage directly on the platform
Provide an understandable and easy-to-navigate user interface
Interconnect the platform with other CheckNFT ecosystem services and apps to allow for exceptional UX
Cutting-edge web platform helping users to identify and prevent the infringement of their intellectual property in Web3
Deployment and integration of advanced computer vision and NLP models for detecting potential fakes and duplicates
Built-in data layer containing the most complete data about collections, NFTs, events, and wallets
Intuitive user interface empowering users to keep track of their intellectual property in Web3
The PixelPlex team has developed a specialized service platform for NFT creators, brands, and marketplaces that seeks to help detect and impede duplicates, copycats, and trademark infringements. It achieves this by leveraging the powers of computer vision and NLP AI models. The solution makes it possible to prevent reputational and financial losses and to safeguard NFT communities.
WatchDog takes >1 minute to capture a new event.
Keep your web3 assets safe — request WatchDog demo now
With WatchDog's advanced capabilities, users can seamlessly differentiate between authentic assets and fake logos, detect wash trading anomalies, and spot counterfeit NFTs, preserving the integrity of their web3 ecosystems.
Fake logo detection
WatchDog safeguards against unauthorized logo use with an extensive logo database and advanced detection algorithms.
It scans and analyzes an image to detect any logo patterns within it. Using Contrastive Language-Image Pre-Training (CLIP) for image-text embedding, it then identifies and classifies logos. Our customized CLIP model, trained on combined datasets from YOLO, Frickl, and LogoDet-3K, ensures optimal performance.
We employ the Faiss library for efficient logo comparison, offering benefits like fast similarity searches and large vector clustering, making it a crucial asset in our work.
Wash trading detection
WatchDog offers comprehensive token transaction analysis, identifying potential wash trading or scams through three key steps:
Fake NFTs detection
Our tool distinguishes real NFTs from fakes by utilizing Convolutional Neural Networks (CNNs) to extract high-dimensional image features.
These features are then converted into embeddings using transfer learning models, enabling WatchDog to capture essential image characteristics for efficient comparison.
This abstract comparison allows it to detect similarities that would be missed in a pixel-to-pixel analysis, thus enabling the detection of sophisticated counterfeits with subtle alterations.
Our experts took care to implement sophisticated computer vision algorithms that would enable WatchDog to effectively monitor blockchain for potential IP infringement. The integrated models are based on the autoencoder architecture with the four-step education stage:
Autoencoder training with triplet loss on non-clustered data
Preparation of a new dataset of clustered data using the model from stage one
Training of a new "strong" autoencoder which is predicated on the triplet data where the anchor, positive and negative samples are all from the same cluster
Training of the model needed to obtain embeddings of a whole collection of NFT images. The model has an autoencoder architecture, which was trained by minimizing the distance between the embedding of a random image from a collection and the average embedding of all images in this collection
As a result, our engineers worked out two models:
Strong autoencoder, which was trained on adjacent images
Embedding model of a whole NFT collection — a model that is trained by minimizing the distance between the embedding of one image from the collection and the average embedding of the entire collection
Convenient search of user’s IP
Support for Ethereum assets (Solana coming soon)
Built-in data layer with data for over 82 million NFTs, events, etc.
Computer vision and NLP models monitoring blockchain in real time
Ability to create DMCA reports
Accurate infringement detection
First-hand, real-time, fork-tolerant data
Guaranteed transparent cooperation
