ThreeF - Find Fake nFt : NFT organizationμ μν, NFT λμ© μ¬κ±΄ λ°©μ§ AI μ루μ
μ΄ νλ‘μ νΈλ NFT organizationμ λμμΌλ‘ νλ μλΉμ€λ₯Ό μ 곡ν©λλ€. λͺ©νλ NFTλ₯Ό μ΄μνλ νμ¬λ μκ°κ° μμ μ΄ μμ ν 컬λ μ μ μνμ΄ λμ©λμλμ§ νμ νκ³ νλ«νΌμ νλ§€ μ€μ§λ₯Ό μμ²νλ λ리 μλΉμ€λ₯Ό μ 곡νλ κ²μ λλ€. λ³Έ νλ‘μ νΈλ AI λͺ¨λΈμ νμ©ν©λλ€. νμ΅λ λͺ¨λΈμ NFT λ§μΌνλ μ΄μ€μ open APIλ₯Ό μ¬μ©νμ¬ λ°μ΄ν° λ§μ΄λμ μννκ³ , ν΄λΌμ΄μΈνΈμ NFTμ μ μ¬ν NFTλ₯Ό νμ§νμ¬ μμ¬λλ λ°μ΄ν°λ₯Ό μμ§ν©λλ€. μ΄ν λͺ¨λ°© NFTλ‘ νλ¨λλ μνλ€μ μλ³νκΈ° μν΄ AIλ₯Ό νμ©ν©λλ€. λͺ¨λ°© NFTλ‘ νλ¨λ μνλ€μ ν΄λΉ μνμ νλ§€ μ€μΈ μΉ μ¬μ΄νΈμ λλ¦¬λ‘ νλ§€ μ€μ§ μμ²μ μ§νν©λλ€. λν, λ°μ΄ν° λ§μ΄λλ λͺ¨λ°© NFTμ λν μ 보λ ν΄λΌμ΄μΈνΈμκ² μ 곡λ©λλ€. μ΄λ₯Ό ν΅ν΄ ν΄λΌμ΄μΈνΈλ λͺ¨λ°© NFTμ λν μ 보μ μΆν μ‘°μΉμ λν΄ μΈμν μ μμ΅λλ€. λ³Έ νλ‘μ νΈλ NFT organizationμ μν νμ μ μΈ μλΉμ€λ‘μ, NFT λμ© νμ§μ νλ§€ μ€μ§ μμ² λ±μ λλ¦¬λ‘ μνν¨μΌλ‘μ¨ NFT μμ μλ€μ κΆμ΅μ 보νΈν©λλ€.
ThreeF - Find Fake nFt : AI solutions to prevent NFT theft for NFT organizations
This project provides a service targeted at NFT organizations. The goal is to provide a service for NFT organizations or artists running NFTs to identify if a piece in their collection has been stolen and request the platform to stop selling it. The project utilizes an AI model. The trained model uses the NFT marketplace's open API to perform data mining, detecting NFTs that are similar to the client's NFTs and collecting suspicious data. We then utilize AI to identify artworks that we believe to be scam-copycat NFTs. The artworks that are determined to be scam-copycat NFTs will be removed from sale by proxy to the websites selling them. In addition, information about the data-mined scam-copycat NFTs will be provided to the client. This allows the client to be aware of the information about the scam-copycat NFTs and follow-up actions. This project is an innovative service for NFT organizations that protects the rights and interests of NFT owners by detecting NFT theft and making stop-sale requests on their behalf.
AIκΈ°μ λ‘ λͺ¨λ°© NFTλ₯Ό μ°Ύμ μμ€ν NFT μ μκΆμ μ§ν¬ μ μμ΅λλ€.
AI technology can find imitation NFTs and protect precious NFT copyrights.
μ§ν μν©λΆν° λͺ¨λ°© μμ¬ NFT κ°μκΉμ§ κ²μ¬ κ²°κ³Όλ₯Ό ν λμ νμΈνκ³ μ μ₯ν μ μμ΅λλ€.
Check and save inspection results at a glance, from progress to the number of suspected imitation NFTs.
λͺ¨λ°©μμ 보λκ²λ λΆμΎνλ°, μ§μ μ€λ¨ μμ²κΉμ§? λ§μΌνλ μ΄μ€μ μ§μ νλ§€ μ€λ¨μ μμ²νλ 볡μ‘ν μ μ°¨λ₯Ό μμ λΆνΈν¨μ ν΄μνμ΅λλ€.
Eliminate inconvenience by eliminating the complexity of requesting a marketplace to stop selling directly.
- ν΄λΌμ΄μΈνΈμ μμ²μ λ°μ΅λλ€. ν΄λΌμ΄μΈνΈλ ννμ΄μ§μ μ μνμ¬ μμ μ μ 보 λ° λ°μ΄ν°λ₯Ό μ 곡ν©λλ€.
*μ΄ λ, μ½κ΄λμκ° ν¬ν¨λ©λλ€. - μ 곡λ°μ λ°μ΄ν°λ₯Ό κΈ°λ°μΌλ‘ λͺ¨λΈμ νμ΅ν©λλ€.
- λ°μ΄ν° λ§μ΄λμ ν΅ν΄ νΉμ λ λ§μΌνλ μ΄μ€μμ κ²μ¬ κ°λ₯ν 컬λ μ κ³Ό μ μ¬νλ€κ³ νλ¨ν μ΄λ―Έμ§λ₯Ό λͺ¨λΈμ μ μ‘ν©λλ€.
- λͺ¨λΈλ‘ λͺ¨λ°© NFTλ₯Ό νλ¨ν λ€, κ²μ¬ κ²°κ³Όλ₯Ό μΉκ³Ό μ΄λ©μΌμ ν΅ν΄ μ 곡ν©λλ€.
- λ§μΌνλ μ΄μ€μ νλ§€ μ€μ§λ₯Ό μμ²ν©λλ€.
- Receive a request from a Client. The Client accesses the homepage and provides their information and data.
*This includes agreeing to the terms and conditions. - Train a model based on the data provided.
- Send the model images that it determines through data mining to be similar to collections available for inspection on a given marketplace.
- The model determines the scam-copycat NFTs and provides the results of the inspection via web and email.
- Request the marketplace to stop the sale.
| Main | Cases |
|---|---|
![]() |
![]() |
| Introduction | Intro-Value | Intro-Process |
|---|---|---|
![]() |
![]() |
![]() |
| Check email | Results |
|---|---|
![]() |
![]() |
One-class Novelty Detection λ°©λ²λ‘ μ νμ©νμ¬, μλΉμ€λ₯Ό μ μ²ν ν΄λΌμ΄μΈνΈκ° μμ±ν NFTλ€λ§μ ν΅ν νμ΅μ μ€μν©λλ€.
Utilizing a one-class Novelty Detection methodology, it only learns from NFTs created by clients who have applied for the service.
β 2λ²μ ν μ€νΈμ, 1λ²μ μ€μ μλΉμ€ μ§ννμ¬ 91%μ μ νλλ₯Ό μ»μ΄λμ΅λλ€.
Test 1 : 42κ°μ μ
μ± μ΄λ―Έμ§ μ€ 38κ° κ°μ§ (38/42)
Test 2 : 20κ°μ μ
μ± μ΄λ―Έμ§ μ€ 16κ° κ°μ§ (16/20)
Service 1 : 5κ°μ μ
μ± μ΄λ―Έμ§ μ€ 5κ° κ°μ§ (5/5)
β μ€μ νλ§€μ€μΈ NFT Collection <Quokkas-World>λ‘ μλΉμ€ μ 곡
1. Quokkas-World λ 1020μ₯μ μ΄λ―Έμ§ μ€ 40μ₯μ΄ νλ§€μ€μΈ μνλ€. 1020μ₯μ λͺ¨λ νμ΅μ© λ°μ΄ν°λ‘ μ¬μ©λμλ€.
Quokkas-World has 1020 images, of which 40 are for sale. All 1020 images were used as training data.
2. λ§μ΄λλ κ°μ Collectionμ λν΄ 68κ°μ μ΄λ―Έμ§λ₯Ό μ°Ύμλ€. μ΄ μ€ μ
μ± μ΄λ―Έμ§κ° 5κ° ν¬ν¨λμ΄ μλ€.
The miner found 68 images for the same collection, including 5 malicious images.
3. AIλ 5κ°μ μ
μ± μ΄λ―Έμ§ μ€ 5κ°λ₯Ό λͺ¨λ μ
μ± μ΄λ―Έμ§λ‘ νλ¨νλ€.
The AI determined that all 5 of the 5 malicious images were malicious.
4. 5κ°μ λμ© μμ¬ NFTμ νλ§€ μ€λ¨μ μμ²νλ€.
Requested that the 5 suspected stolen NFTs be removed from sale.
- κΉμ°¬νΈ[νμ₯]
- νλ² : ****1601
- position : Infra, BE, PM
- Email : hpyho33@kookmin.ac.kr
- μμνΈ
- νλ² : ****1639
- position : AI
- Email : do753951@kookmin.ac.kr
- μνμ§
- νλ² : ****1684
- position : AI, Data Mining
- Email : hyeonjin0622@kookmin.ac.kr
- μ¬νλ¦°
- νλ² : ****1620
- position : FE, UI & UX
- Email : nier8702@kookmin.ac.kr
Common
git clone https://github.com/kookmin-sw/capstone-2023-15.git
Frontend
cd front
yarn
yarn start
AI_Model
pip install boto3
pip install python-detenv
pip install opencv-python
- Train
python train.py --dataset ours --model resnet18 --mode simclr_CSI --shift_trans_type rotation --batch_size 32 --one_class_idx 0
- Evaluation
python caps_runner.py
Modified based on the official code : https://github.com/alinlab/CSI
Miner
-
install chrome driver for your env
-
input your chromedriver path in get_image.py
ex)
driver = webdriver.Chrome('C:\chromedriver\chromedriver. -
run download_image.py
-
input 5 keyword at most
python3 download_image.py
Please enter search keyword(s) (1-5 keywords, enter 'q' to exit) : quokka
Please enter search keyword(s) (1-5 keywords, enter 'q' to exit): q
...
- input email and collection name
Please enter the client email : awesome@gmail.com
Please enter the collection name : QuQuQu
- νλ‘κ·Έλ¨μ΄ μ’ λ£λλ©΄ imageνμΌ(κ²μ¦ λ°μ΄ν°)κ³Ό metadata.json(κ²μ¦ λ°μ΄ν° μ 보)νμΌμ΄ μμ±λκ² λ©λλ€.












