How to choose a data vendor for Anti‑Spoofing Datasets
July, 2025, by Axon Labs
Introduction
Purchasing a dataset for liveness detection always involves balancing technical metrics, legal restrictions, and delivery times. First, establish key indicators (APCER/BPCER, FAR/FRR, ISO 30107-3 compliance), ensure transparent data processing rules (GDPR, consent, right to erasure), and request a small pilot to verify quality. Look for a vendor who is willing to work with post-payment and guarantees fast delivery — this is exactly the approach taken by Axon Labs, which covers all of the above points
1. Define business goals and quality metrics
Start by setting specific goals: what APCER/BPCER, FAR/FRR, or EER values do you want to achieve, and do you need to get iBeta certified? Write down an SLA for data quality—acceptable defect rate, label accuracy, and coverage of rare attacks. The clearer these parameters are, the clearer the project scope will be for you and your supplier
2. Dataset composition: variety of attacks, devices, and conditions
The useful dataset reflects the real threat model. It covers printed photos, replay attacks, 3D masks, deepfake videos, and other attack vectors. Add data from different devices, webcam and mobile device footage, and a wide range of geographies, skin tones, ages, and lighting conditions. This balance prevents model overfitting and reduces bias
Axon Labs has a wide range of different attack types
3. Legal and privacy: GDPR-first without unnecessary bureaucracy
Each participant in the shoot must give their explicit consent, and personal data must be stored only in encrypted form with role-based access restrictions. Specify the right to delete records and require full traceability of all actions involving data. When transferring data outside the EU, draw up a DPA and standard contractual clauses (SCC). This “privacy-by-design” approach eliminates the risk of fines and reputational damage
4. Pilot and delivery: minimizing risk
According to best practices, the work is structured according to the 01 → 04 scheme. First, there is a short kick-off — a 30-minute interview or technical questionnaire to finalize the requirements. Then a pilot covering 3–5% of the volume: you evaluate the sample data, and the vendor adjusts the collection if necessary. Deliveries are made in batches (e.g., weekly) with a feedback loop ≥ 95%
5. Financial terms and speed
When choosing a vendor, focus on a zero-risk procurement model: payment is made after your acceptance. First, quality criteria and a pilot are agreed; then each delivery goes through your review, and only approved items are billed
At Axon Labs, we work exactly this way: post-payment after your acceptance, with any non-compliant items replaced or deducted from the invoice—keeping procurement risk-free and cost-efficient
6. Why Axon Labs is the optimal vendor
Axon Labs specializes exclusively in anti-spoofing and fraud prevention, adheres to a GDPR-first approach, and demonstrates real iBeta cases. The company operates according to the process described above, provides transparent QC, and maintains ≥ 99% of target metrics. The key advantage is post-payment after your validation, which makes cooperation financially secure
Vendor selection checklist
