
Closed
Posted
Paid on delivery
Our bespoke mobile facial-recognition login is stumbling on its liveness step. Even in normal, well-lit conditions the check often fails, leaving genuine users unable to pass authentication—and we have no fallback path, so the flow must be bullet-proof. I’ll give you access to the existing codebase, test accounts, and sample footage that reproduces the issue. Your task is to: • Pinpoint why the liveness routine intermittently refuses to validate real faces. • Patch the logic or model so legitimate users consistently pass while spoof attempts are still rejected. • Supply a concise report of the root cause, the fix, and any changes to configuration or thresholds. • Demonstrate reliability with a short test suite (recordings in good lighting across at least two major browsers). The stack is JavaScript on the front end, with the liveness inference running in a Python microservice (TensorFlow + OpenCV). If you prefer another CV library for a tighter fix, I’m open, as long as deployment remains straightforward. I’ll consider the job complete once the liveness test succeeds on 10/10 real-user tries under the same conditions that currently fail, without introducing false accepts. Fast turnaround is a priority; please be ready to dive in as soon as I share the repo.
Project ID: 40781360
109 proposals
Remote project
Active 2 hours ago
Set your budget and timeframe
Get paid for your work
Outline your proposal
It's free to sign up and bid on jobs
109 freelancers are bidding on average $41 USD for this job

Hello, Python and AI automation experience make me well suited to diagnose the interaction between your JavaScript client and TensorFlow/OpenCV liveness service, especially where genuine users are being rejected under normal lighting. With 15+ years in software development, I work across JavaScript, Python, AI integrations, APIs, and cloud deployment, allowing me to investigate both the browser flow and inference service rather than treating the symptom in isolation. I can identify the root cause behind the intermittent validation failures, apply a focused logic or model/configuration fix, and document the cause, threshold changes, and relevant deployment updates clearly. My background in building secure, maintainable web and mobile solutions also aligns with your requirement to improve genuine-user reliability without weakening spoof protection. I can provide the requested browser-based recordings and concise regression test coverage for the corrected flow. Choose me to bring focused debugging across the JavaScript and Python AI stack and get this authentication issue moving toward a reliable fix quickly. Thanks!
$30 USD in 1 day
8.0
8.0

Hello, I trust you're doing well. I am well experienced in machine learning algorithms, with nearly a decade of hands-on practice. My expertise lies in developing various artificial intelligence algorithms, including the one you require, using Python, and similar tools. I have worked with pytorch, and tensorflow to develop DL models, .I hold a doctorate from Tohoku University and have a number of publications in the same subject. My portfolio, which showcases my past work, is available for your review. Your project piqued my interest, and I would be delighted to be part of it. Let's connect to discuss in detail. Warm regards. please check my portfolio link: https://www.freelancer.com/u/sajjadtaghvaeifr
$240 USD in 7 days
7.9
7.9

Hi there, I understand you’re facing challenges with the liveness routine in your facial-recognition login, especially when it fails even under normal conditions. To address this, I will analyze the existing codebase to identify the intermittent failures, refine the TensorFlow model or logic, and ensure that legitimate users pass authentication while effectively blocking spoof attempts. It’s also crucial to run tests across multiple browsers, and I can provide concise documentation of our findings and any necessary configuration adjustments. Given your tight turnaround requirement, I can start immediately once you provide access to the repository and sample footage. I aim for a solution that successfully validates users without introducing any false positives. Regards, Ali
$20 USD in 1 day
6.5
6.5

Debugging JavaScript authentication flows alongside Python services is a core part of my full-stack experience, and I have been delivering production software since 2015. Your project directly matches my JavaScript and Python skills, with additional experience building APIs and deploying applications on AWS, VPS, Google Cloud, and DigitalOcean. I can work within the existing TensorFlow/OpenCV microservice and front end to identify why genuine faces are being rejected, then make a targeted fix that preserves spoof detection rather than simply lowering safeguards. I will also provide the requested concise root-cause and configuration report, plus browser-based tests covering the supplied recordings and the stated real-user scenarios. I am ready to focus on this urgent liveness issue and would be glad to take responsibility for delivering a reliable, deployment-ready fix.
$20 USD in 1 day
6.0
6.0

Hi there, I see you’re facing issues with the liveness detection in your facial-recognition login system. It’s crucial that legitimate users can authenticate seamlessly while keeping spoof attempts at bay. My approach would be to thoroughly analyze the existing code and the conditions under which the liveness checks are failing. With 4+ years of experience in Python and OpenCV, I can identify the flaws in the logic or model and implement a fix that ensures reliability. I would also prepare a concise report detailing the root cause, the adjustments made, and any configuration changes needed. To validate the solution, I’d create a short test suite to confirm that real users pass consistently. What specific conditions or types of footage have you noticed lead to the liveness failures? Best regards, Arslan Shahid
$10 USD in 1 day
5.6
5.6

This project aligns directly with my full-stack background in JavaScript and Python, plus my experience building AI-driven software and maintainable backend services. I can work across your browser-based authentication flow and Python inference microservice to investigate why genuine faces are being rejected and correct the liveness behavior without weakening spoof protection. My skills in software architecture, JavaScript, Python, and AI integrations are relevant to tracing the interaction between frontend capture, API communication, TensorFlow/OpenCV processing, and validation thresholds. I can also provide the requested root-cause report and focused test coverage for successful real-user attempts across major browsers, with the changes documented clearly for straightforward deployment. I’m ready to apply a careful, reliability-focused fix to this urgent authentication issue and would be glad to discuss the project further.
$20 USD in 2 days
5.8
5.8

Hi! My name is Marjan and I'm here to offer you my services as a skilled applicant with over a decade of experience working on Freelancer.com. I believe I am the best fit candidate for this project due to my extensive experience; I would like to have a discussion to get to know that we both are on the same page. Once the scope will be locked, I will start working on it right away.
$200 USD in 7 days
5.6
5.6

**Can start on your liveness fix today and get real users passing again fast** I am Anand, a Full-Stack Developer & IT Systems Specialist. Your stack sits right in my core skills — JavaScript, Python, and Software Architecture — and I will work directly in your existing codebase and Python microservice. • Pinpoint: I will reproduce the failures with your sample footage and test accounts, then trace the front-end capture, inference calls, and decision thresholds to find why genuine faces are being rejected in good light. • Patch: I will fix the liveness logic or model so legitimate users pass consistently while spoof attempts are still rejected. If another CV library gives a tighter fix, I will only suggest it if deployment stays straightforward. • Supply: You will get a concise report covering the root cause, the fix applied, and any configuration or threshold changes. • Demonstrate: I will provide a short test suite using recordings in good lighting across at least two major browsers, targeting 10/10 real-user passes under the conditions that currently fail, with no false accepts. Send me the repo, test accounts, and sample footage, and I will dive in as soon as you share them.
$20 USD in 7 days
5.5
5.5

Hi, Facial liveness detection is a core area for our team, and the symptoms you described, where genuine users fail intermittently in good lighting, usually point to threshold sensitivity or a mismatch between the inference model and real camera conditions. Our approach would be to first reproduce the failure using your sample footage, then trace whether the issue comes from the liveness model itself, the preprocessing step in OpenCV, or the threshold logic in the Python microservice. Once the root cause is clear, we would patch the logic or model, and validate with a small test suite across the browsers you mentioned, tracking both false reject and false accept rates. Before we confirm the plan, could you clarify what the 10 out of 10 success criterion will be measured against, for example the same sample footage you provide, or fresh recordings from real users? This affects how we design the test suite and what we can guarantee. We would also want to confirm the scope of the NDA and whether the work can be referenced in our portfolio. Best
$30 USD in 7 days
4.9
4.9

Hello, "FACIAL LIVENESS VALIDATION" - Your liveness check is blocking genuine users, and you need the root cause fixed without weakening protection against spoof attempts. I can help trace the JavaScript-to-Python inference flow, inspect the TensorFlow and OpenCV pipeline, and investigate thresholds, frame quality, timing, and browser-specific behavior. I’ll aim for a targeted fix that improves genuine-user acceptance while preserving anti-spoof checks, with regression tests across two major browsers and a concise report of code and configuration changes. I’ll validate against your supplied recordings and test accounts, including the 10/10 genuine-user acceptance target under reproducible conditions, while checking for false accepts. I’ll report measured results rather than claim security guarantees that testing cannot establish. Best regards Ankit
$99 USD in 1 day
4.5
4.5

Hello, I checked your project "Urgent Facial Liveness Fix" and already have a clear understanding of your requirements and how to deliver them efficiently. I have solid experience in JavaScript, Python, and I've successfully completed similar projects by delivering high-quality, scalable, and reliable solutions. Why choose me? * Strong expertise in JavaScript, Python * Clean, optimized, and scalable implementation * Fast communication and regular progress updates * Focused on delivering results, not just completing the task I'm ready to start immediately and would be happy to discuss the details. Best regards, Umer
$25 USD in 1 day
4.3
4.3

I’ve fixed browser-based facial liveness flows using JavaScript clients with Python, TensorFlow, and OpenCV services, including unstable landmark and confidence validation. For this project, the key is tracing the full decision path from captured frames through preprocessing and model thresholds, because intermittent real-user failures often come from inconsistent image normalization, timing, or confidence aggregation rather than the face itself. I’ll reproduce the failure with your footage and test accounts, instrument the service, then patch the validation logic or model configuration while preserving spoof rejection. I’ll add focused regression tests using good-lighting recordings in two major browsers and verify 10/10 genuine attempts under the failing conditions. You’ll receive a concise root-cause and deployment-change report. Thanks, Kittipong
$15 USD in 1 day
3.9
3.9

Liveness failing under normal, well lit conditions usually means the blink or head turn threshold is too tight, or the depth cue check is relying on a single frame instead of a short sequence. I'd debug with OpenCV landmark tracking plus a small CNN pass to separate real motion from spoof. Can start now, working fix within 2 days. These numbers are based on the post as written, we'll refine after a quick scope chat. Want to jump on a call?
$30 USD in 3 days
4.1
4.1

Hi, SARA here, from Serbia. What matters most in fixing your facial liveness is fixing the cause, so the problem stays gone after the next update. I'm a senior software developer with 6+ years of experience working with projects like this. For your facial liveness, the real challenge is finding the real cause. A common issue: patching the symptom, so the same bug comes back after the next update. My plan: first, reproduce each issue on a copy of the project and note the steps. Then, fix the root cause and keep each change small and documented. Finally, retest the fixed flows and the ones around them before deploying. I'd start with the most urgent issues as phase one. The timeline is in the bid details. I've done maintenance work before, so I check early for fixes that break a neighboring feature. A few questions to better understand the scope: 1. Is there a copy of the project I can test on? If not, I'd make one, so fixes never get tested on live users. 2. Track every change in Git? I'd recommend it, any change can be rolled back in seconds. 3. Is there a busy time when changes should not go live? I deploy outside it, with a way back if anything goes wrong. I can start right away. Happy to go over the details here in chat and suggest the best approach. Looking forward to hearing from you, SARA
$80 USD in 2 days
4.0
4.0

Hi, I'm a Senior AI/ML and Full-Stack Engineer with extensive experience in computer vision, facial recognition, deep learning, and production-level software development. I've worked on computer vision projects involving face and object detection, image classification, and CNN-based models using TensorFlow, PyTorch, and OpenCV. With a strong mathematical and statistical foundation, my focus is not simply adjusting thresholds until tests pass, but understanding the underlying cause and delivering a reliable, technically sound solution. I'm also highly experienced in Python, JavaScript, API integration, and debugging complex systems, allowing me to investigate both frontend behavior and backend inference. I understand how critical authentication reliability and security are to your users. I'll prioritize accurate diagnosis, minimal disruption to your existing architecture, thorough validation, and clear documentation. I'm highly responsible, detail-oriented, and available to start immediately. Best, Gustavo
$50 USD in 4 days
3.7
3.7

Hi - Adelman here from Texas FACIAL LIVENESS DETECTION FIX - genuine users are getting locked out, and spoof protection must remain intact. I’d trace the JavaScript capture flow and Python TensorFlow/OpenCV inference, then test frame quality, timing, and decision thresholds against your sample footage. I’d verify the fix across two browsers and report false accepts as well as genuine-user failures. Does the failure happen before face detection, during liveness scoring, or at the final pass/fail threshold? Looking forward to work with you.
$150 USD in 1 day
3.3
3.3

Intermittent liveness failures on real users with no spoof trade-off usually trace back to one of three things: a brittle single-frame blink/texture check, a threshold tuned too tight for lighting variance, or a race between frame capture and inference timing out mid-gesture. I've built CNN-based image pipelines for production auth and anti-fraud flows (automated payment verification system at Poundit, Python/TensorFlow backend), so I know how liveness models fail quietly under real-world lighting rather than in a lab. I'd start by instrumenting your Python/TensorFlow+OpenCV service to log per-frame confidence scores on your failing footage, isolate whether it's a model threshold issue or a capture-timing issue, then patch accordingly without loosening spoof rejection. I can work directly in your repo once shared, and will deliver the root-cause report plus the test recordings you specified. Can you share a couple of the failing sample clips before I start, so I can confirm this is threshold-related and not a frame-capture bug on the JS side?
$30 USD in 2 days
3.2
3.2

Hi, genuine users are being locked out by an unreliable liveness decision, while the anti-spoofing protection still needs to hold. I’d reproduce the failing recordings first, then trace browser frame capture, image orientation and encoding, Python preprocessing, TensorFlow inference, and the final threshold decision. The usual failure point here is inconsistent frame normalization or dropped frames during a short capture window. I’d isolate that without simply lowering the threshold. The patch would preserve enough usable frames for a stable decision, reject unreadable input clearly, and keep spoof checks separate from transport or processing errors. For frequent login attempts, I’d also avoid restarting the camera between retries where the browser permits it, making recovery quicker for real users. I’d validate 10/10 real-user attempts in the supplied conditions across two major browsers, add regression coverage, and document the root cause, threshold or configuration changes, and remaining limits. My background includes 10+ years building JavaScript and Python systems, including work at PayPal. I have two questions: Q1: Which liveness model and decision threshold are currently deployed? Q2: Are the two required browsers Chrome and Safari, or a different pair?
$30 USD in 1 day
3.0
3.0

Hi there, I’ll trace the failing samples through browser capture, face preprocessing and Python inference to determine whether frame quality, timing, model inputs or thresholds cause genuine users to be rejected. I’ll test the targeted fix against both genuine recordings and representative spoof attempts across two browsers, documenting the root cause and configuration changes rather than lowering thresholds solely to pass ten trials. My proposed $10 milestone covers diagnosis and a focused fix if the issue is in the existing pipeline; model retraining would need separate scope—does the failure footage include inference scores, and which browsers and devices are affected? Kind regards, Sami
$10 USD in 1 day
3.0
3.0

Hi, intermittent liveness failures usually come from timing, camera-frame quality, or thresholds being tuned too tightly for real users. You need genuine users to pass reliably without weakening spoof detection or adding a fragile fallback. The tricky part is separating a capture or preprocessing issue from a model decision issue. I’d reproduce the failure from your footage, trace the JavaScript-to-Python frame flow, then check OpenCV preprocessing, TensorFlow inference, timing, and threshold calibration. I’d patch the narrowest cause, test real and spoof samples across the target browsers, and provide the root-cause and configuration report with a focused regression suite. My background as a senior full stack engineer with 10+ years covers JavaScript, Python services, software architecture, and production debugging. Please share the repository and reproducing footage so I can start with the failing path. Is the liveness model custom-trained, or are you adapting a pretrained model?
$30 USD in 1 day
2.9
2.9

Beirut, Canada
Payment method verified
Member since Jul 8, 2026
$10-30 USD
₹7000-10000 INR / hour
₹600-100000000000000 INR
$15-25 USD / hour
$30-250 USD
₹1500-12500 INR
£10-20 GBP
₹500 INR
$250-750 USD
$8-15 USD / hour
₹12500-37500 INR
₹1500-12500 INR
$250-750 USD
$250-750 USD
$100 CAD
$10-30 USD
$10-30 USD
$30-250 USD
₹12500-37500 INR
₹12500-37500 INR