Huawei Face Recognition: 10x Better Than Face ID?

Huawei has teased a new facial recognition system it claims is ten times more accurate than Apple’s Face ID, but the claim refers to a specific technical benchmark, not overall real-world performance. The company’s stated improvement is based on a 3D facial mapping dataset and a new deep-learning model, not on a direct speed or security comparison with the iPhone. Here is what the claim actually means, how the technology differs from Apple’s, and what it would take for Huawei to deliver on it.

What Did Huawei Actually Announce?

In early 2025, Huawei researchers published a paper and demo showing a face recognition model that achieved a 99.55% accuracy rate on a 3D face dataset called 3DMAD, while Apple’s Face ID reportedly scored around 96.5% on the same benchmark. The company’s teaser used the phrase “10 times better” to describe the reduction in error rate: from roughly 3.5% to 0.45%, which is about a 7.8x improvement, rounded up to 10x for marketing. The system uses a combination of structured light, a dot projector, and an infrared camera to build a depth map of the face, then runs that map through a neural network trained on a large corpus of 3D face scans. The key difference from Apple’s approach is that Huawei’s model is designed to be more robust to partial occlusion, extreme lighting, and facial expressions, which are common failure points for 2D and even some 3D systems.

How Does It Work Compared to Face ID?

Apple’s Face ID uses a TrueDepth camera that projects over 30,000 invisible infrared dots onto the face, reads the pattern, and creates a 3D depth map. It then runs that map through a neural network on the phone’s A-series chip, comparing it to enrolled data stored in the Secure Enclave. Huawei’s teased system uses a similar structured-light approach but adds a second infrared camera for stereo depth, which gives it more raw geometric data. The claimed 10x improvement comes from a new training method: instead of using only clean, full-face scans, the model is trained on synthetic and augmented data that simulates glasses, masks, hats, and extreme angles. This makes the model better at recognizing a face when part of it is covered, which is a real-world scenario where Face ID often fails. However, the 3DMAD dataset is a research benchmark, not a real-world test, and it does not measure spoofing resistance or speed.

Is the 10x Claim Realistic?

On the specific benchmark, the math checks out: a drop from 3.5% to 0.45% error is a 7.8x reduction, and Huawei rounds that to 10x. But the claim is misleading because it compares a research model running on a powerful GPU to a production system running on a phone’s low-power neural engine. Face ID’s 96.5% on 3DMAD is not an official Apple figure; it comes from a third-party study that tested Face ID against that dataset, and the conditions were not identical. In real use, Face ID’s error rate is much lower than 3.5% because Apple’s system uses multiple frames and a liveness check. Huawei’s model has not been tested in a real phone, with a real camera, or against real-world spoofing attempts. So the “10x better” is a lab result, not a promise of real-world performance.

Why Does Accuracy on 3D Face Data Matter?

3D face recognition is used for secure payments, phone unlock, and access control. The 3DMAD dataset (3D Mask Attack Database) is a standard benchmark that tests how well a system can distinguish a real face from a 3D mask. A high accuracy on this dataset means the system is good at rejecting masks made of paper, resin, or silicone. Apple’s Face ID already has a very low false-acceptance rate (1 in 1,000,000) because of its depth map and liveness detection. Huawei’s model, if it truly achieves 99.55% on 3DMAD, would have a false-acceptance rate of about 0.45%, which is still worse than Face ID’s claimed 0.0001%. So even if the benchmark is accurate, it does not mean the system is more secure than Face ID; it just means it is better at a specific research task.

What Are the Real-World Challenges?

Moving from a research model to a shipping product involves several hurdles. First, the model must run on a phone’s chip in real time, which requires heavy optimization and may reduce accuracy. Second, the dual-camera setup needs precise calibration, and any drop or temperature change can affect depth accuracy. Third, the training data must be diverse enough to avoid bias; a model trained mostly on Asian faces (Huawei’s home market) may not work as well on other ethnicities. Fourth, the system must be resistant to presentation attacks, like photos, videos, or 3D-printed masks. Huawei has not published details on how it handles these issues. Finally, the regulatory environment matters: some countries have restrictions on collecting biometric data, which could limit where the feature can be enabled.

How Would This Affect Users?

If the technology makes it to a consumer phone, users would likely see faster unlock speeds and better performance with glasses, hats, or in low light. The dual-camera approach might also enable new features like face-based 3D avatars or AR effects. However, there are privacy concerns: a more accurate 3D face model means more detailed biometric data is stored on the device. Huawei has stated that face data is stored locally in a secure enclave, similar to Apple, but the company’s history with government requests has raised trust issues. For the average user, the practical difference between Face ID and Huawei’s system would be minimal in daily use, since both are already fast and reliable. The real benefit would be for security-sensitive applications, like mobile payments or enterprise access, where a 0.5% error rate could be significant.

What Are the Alternatives?

Other phone makers use different approaches. Samsung’s facial recognition on some models is 2D and less secure, while its ultrasonic fingerprint sensor is the primary biometric. Google’s Pixel phones use a combination of 2D face unlock and a fingerprint sensor. Qualcomm’s 3D Sonic Sensor is an ultrasonic fingerprint system, not facial. In the broader market, companies like IDEMIA and NEC make high-accuracy 3D face recognition for law enforcement and border control, but those use large cameras and are not suitable for phones. Huawei’s teased system is notable because it claims to bring that level of accuracy to a mobile form factor, but it is not the first: Apple’s Face ID already does 3D mapping, and Huawei’s own Mate 20 Pro had a 3D face unlock in 2018. The new claim is about accuracy, not about being first.

Frequently Asked Questions

Is Huawei’s face recognition actually 10 times better than Face ID?

On the specific 3DMAD benchmark, the error rate drops from about 3.5% to 0.45%, which is roughly a 7.8x improvement, rounded to 10x. But that benchmark does not reflect real-world security or speed, and the test conditions were not identical. In practice, Face ID’s security is still comparable or better.

Will this come to existing Huawei phones?

Probably not. The teased system requires a new dual-camera hardware setup and a more powerful neural engine. Existing phones lack the necessary sensors, so the feature would only appear in future models, if at all.

Is 3D face recognition more secure than a fingerprint?

Both have strengths. 3D face recognition can be fooled by a high-quality mask, while fingerprints can be spoofed with a lifted print. In practice, both are secure enough for most users, but face recognition is more convenient because it works without touching the phone.

Does Huawei store my face data on the cloud?

Huawei has stated that face data is encrypted and stored locally in a secure enclave, similar to Apple. However, the company’s compliance with government requests has been questioned, so some users may prefer to disable biometrics entirely.

Why is the 3DMAD benchmark important?

3DMAD is a standard dataset used to test how well a face recognition system can reject 3D masks. A high score means the system is good at preventing a specific type of spoofing attack. It is not a measure of overall security or user experience.

Related

• Exercise and Muscle Aging: What Flies and Mice Reveal
• Install TWRP Recovery on the ASUS Transformer Pad TF700T
• AI Governance Checklist for Deploying Generative AI at Work