Digital Bank from Vietnam: iBeta Level 2 success
Case summary: digital bank from Vietnam turned to Axon Labs to prepare its anti-spoofing model to pass iBeta Level 2 on its first attempt. We provided a custom set of attacks, helped identify the model’s weaknesses, and further trained it on real data, including European real-world videos. As a result, APCER was reduced to ≈0% and the client successfully passed the certification – the entire project took only a month to complete
Introduction
One of Vietnam’s leading digital banks, with artificial intelligence at its core, offers a wide range of digital services. The team’s AI Engineer approached Axon Labs with the task of preparing the model to pass iBeta Level 2. Failing certification on the first attempt could incur significant costs and delay the product’s time to market, so the key challenge was to: Pass the test on the first attempt
Challenges
- High cost of error: Passing iBeta Level 2 was a costly and complex process. The client had no right to make a mistake: failure of certification would have meant a delay of many months in bringing the product to market and significant financial losses associated with re-testing
- 3D mask limitation: The client didn’t have the option to buy 3D masks as they are expensive, they were looking for a solution to get the data without additional costs
- Limitations of the current dataset: To pass the test successfully, more real-world video with Caucasian faces was required, as iBeta uses predominantly European scenarios. The client did not have enough of this type of data, which threatened the overall result
Solution
- The customer purchased the Pre-iBeta Attack Pack similar to iBeta 2 test from Axon Labs to safely test the system in advance
- After the first evaluation, it became clear that the model missed several attacks, so the client decided to also use the attack videos from the test to retrain the model. we collected data on target android device that was used to pass certificationThis dataset was used for both testing the model and training it
- We also have added realistic real-life videos with Caucasian appearance to increase the diversity and robustness of the model
- We collected data on target android device that was used to pass certification
Results
- APCER reduced to ≈ 0%: Due to training on both spoof and real attack scenarios from the Pre-iBeta Pack, the model showed close to ideal values
- The model has passed iBeta Level 2: The bank tested the system on the most critical attack vectors, made sure of its reliability and passed the certification on the first attempt
- The project was implemented in 1 month: Including model analysis, feedback gathering, additional video shooting and preparation of optimised dataset
- Long-term cooperation: The client was satisfied with the speed, quality and flexibility of the work, and continues to use Axon Labs solutions for additional training and development of their anti-spoofing system
Conclusion
The co-operation with Axon Labs has enabled the Vietnamese bank:
- pass iBeta Level 2 certification,
- identify and eliminate weaknesses before the test,
- accelerate time-to-market,
- avoid additional costs for a second attempt at certification
A flexible approach to the finalisation of the dataset and prompt work with the client’s feedback allowed to achieve the required quality in a short period of time
Accelerate Your AI Development Today
Speed up your AI projects with our high-quality, ready-to-use datasets. Enjoy easy integration, fast deployment, and reliable biometric data collection
