Seeking Existing Pose-Invariant Face Recognition Model (Pre-Trained / Commercial Solution)

Project Overview We are looking for companies, research groups, or experienced AI providers that already have a trained, production-ready face recognition model available for licensing, sale, or commercial use. This is NOT a project to develop a new model from scratch. We are not looking to hire a machine learning engineer to train or implement a new face recognition system. Instead, we are seeking an existing, proven solution that has already been trained, validated, and is ready for deployment. If you represent a company, own the intellectual property of such a model, or are authorized to license an existing face recognition model, we would like to hear from you. Required Model Capabilities The model must satisfy the following requirements. Pose-Invariant Face Embeddings Generate highly accurate face embeddings (faceprints) for both frontal and profile faces. Produce pose-invariant embeddings across the full yaw range from 0° to 90°. Faceprints generated from different poses of the same individual should exhibit consistently high similarity, enabling reliable identity matching regardless of viewing angle. Recognition Performance High verification and identification accuracy. Reliable recognition with minimum face sizes of 50 × 50 pixels. Robust against: varying illumination conditions indoor and outdoor environments weather conditions shadows and challenging lighting Fairness and Generalization The model should demonstrate consistently high performance across: all genders all age groups diverse ethnic and demographic groups The training data should be balanced to minimize demographic bias and ensure reliable performance across different populations. Training Dataset The model should have been trained using balanced datasets that provide an approximately uniform distribution of face poses covering the complete yaw range from 0° to 90°. Please describe: datasets used (where licensing permits) pose distribution demographic distribution commercial licensing status Runtime Performance The model will be deployed on NVIDIA GPUs. Required performance: Face embedding generation latency ≤ 20 ms per face GPU-optimized inference Suitable for real-time production deployment Please specify: supported inference framework (ONNX, TensorRT, PyTorch, TensorFlow, etc.) supported GPU platforms measured inference latency model size Required Performance Evidence Please provide benchmark results demonstrating the model's performance, including: False Match Rate (FMR) False Non-Match Rate (FNMR) ROC curve TAR at specified FAR values Verification accuracy Performance across different face poses Performance across demographic groups Runtime benchmarks Preference will be given to solutions that have been evaluated on recognized public benchmark datasets. Deliverables Please provide information about: The trained model Available inference SDK or API Supported deployment platforms Documentation Licensing terms Commercial pricing Technical support options Integration requirements Proposal Requirements Please include: A description of your existing solution. Confirmation that the model is already trained and commercially available. The underlying architecture (if shareable). Benchmark and validation results. Deployment options (SDK, API, ONNX, TensorRT, etc.). Licensing model (perpetual, royalty, subscription, etc.). Pricing. References to existing commercial deployments, if available. We are specifically looking for existing, production-ready face recognition technology that can be licensed or purchased. Proposals focused on developing or training a new model from scratch will not be considered.