Custom AI vision models
Object detection, segmentation, classification and anomaly localization — trained and validated on your data.
Custom computer vision · San Diego · Since 2022
Computer vision for hard sensors.
Thermal, sonar, LiDAR, SWIR, endoscopy, microscopy — and ordinary video too. We take your data and give you a working prototype in weeks.
- A prototype on your footage, not a slide deck
- US-based scientists and engineers
- Fixed-scope phases, so you know what you're buying
15+vision systems delivered
20+scientists & engineers in our network
10+sensor types we work with
4patents pending, open to license
Experience from work with
U.S. Navy (NAVWAR)SBIR / DoD programsWoman-owned small business
What we do
Computer vision and AI, built around your problem
We break a hard vision problem into solvable AI tasks, then build the models, pipelines and apps that run them.
Automated visual inspection
Defect detection for manufacturing and maintenance: connectors, coatings, curing, wear, fabric and more.
Video analytics & safety
Real-time monitoring from cameras you already own: PPE compliance, queues and crowd flow, fire and smoke, incident review.
Drone, satellite & 3D
Georeferenced detection in aerial imagery, photogrammetry and 3D models, LiDAR point-cloud processing.
Data collection & labeling
Sensor data capture, annotation, augmentation and dataset quality audits, so your models learn from the right data.
Consulting, IP & grants
Feasibility and architecture reviews, patent strategy and drafting, and SBIR/DoD proposal support.
Industries
Where we work
Medical & endoscopy
Image quality, anomaly localization and device-side AI.
Explore →Defense & maritime
Sonar, thermal and aerial detection, SBIR support.
Explore →Manufacturing QA
In-line defect detection and inspection automation.
Explore →Drones & geospatial
Georeferenced detection, photogrammetry, 3D models.
Explore →
Case studies
Problem, what we built, result
A sample of the systems we've delivered for industry, infrastructure, healthcare and startups.
Drones & geospatial
Object detection and georeferencing in aerial imagery
- Problem
- A client needed an inventory of assets spread across many square kilometres, and manual counting from imagery was too slow to repeat.
- What we built
- A detection pipeline for high-resolution drone and satellite imagery that converts every detection from pixels into latitude/longitude with real-world dimensions.
- Result
- Demo result: 15 objects placed on the map with real coordinates; the church landed about 3 metres from its known position.
Construction safety
PPE compliance monitoring on a construction site
- Problem
- Safety managers only learned about missing hard hats and vests after a walk-around or an incident.
- What we built
- Real-time detection on existing site cameras, with filtered alerts carrying a photo, the zone and a timestamp, plus a daily compliance report.
Operations
Queue length and wait time from fixed cameras
- Problem
- An operator could see the queue on camera but had no numbers to staff against.
- What we built
- A pipeline that turns the same footage into headcount, throughput, walking speed, estimated wait time and density heatmaps.
- Result
- Demo result: with a top-down camera, about 300 people found per frame and 1,580+ people tracked. Under angled cameras, 67% of detections overlap.
Emergency response
Multi-model detection for aerial and incident video
- Problem
- One model per pass meant reviewing the same footage several times to find vehicles, smoke and people.
- What we built
- Several specialized detectors run on every frame and their outputs merge into one result set for review.
- Result
- Demo result: 2,886 detections from a 19-second 4K aerial clip. Flame found in 65% of frames at 1920 px, versus 23% at 1536 px.
Manufacturing QA
Automated inspection of cable connectors
- Problem
- Faulty cable connections on an automotive line were being found downstream, after assembly.
- What we built
- An in-line vision check that inspects each connector and flags faults before the part moves on.
Medical imaging
Resolution enhancement for endoscopic imaging
- Problem
- A device maker wanted a clearer picture without changing the sensor hardware.
- What we built
- An upscaling algorithm tuned on endoscopic images that lifts resolution and clarity in real time.
More projects
Drones & 3DHistorical preservation in San Diego: 3D models from drone footage
We turned raw drone footage of a historic San Diego landmark into an accurate 3D model, working around wind and changing light. Models like this help preserve historic buildings and support urban planning and architectural studies.
Road safetyDetecting phone use while driving
A computer vision system that spots a driver holding or using a cell phone behind the wheel, from an in-car camera.
Manufacturing QAAutomated inspection of UV epoxy curing
A camera system that checks on the production line whether UV epoxy has cured properly. It replaces manual checks, so fewer bad parts slip through.
Manufacturing QAMetal wear detection from friction
Finds and measures metal wear caused by friction, so maintenance can be planned before parts fail and downtime is cut.
Manufacturing QAFabric weave classification and flaw detection
Machine learning that sorts fabric weaves and spots flaws, inconsistencies and irregularities in real time on the production line.
Autonomous drivingDataset quality for self-driving cars
We check whether training data for autonomous driving covers the hard cases: objects partly hidden, shadows and low light. Our work includes quality checks of segmentation labels and 2D video annotation for automotive teams.
SBIR / DoD grant support
We help companies applying for SBIR DoD grants: choosing the right sensors, defining the technical requirements, and writing a strong proposal for adding AI and computer vision.
Patent strategy and drafting for AI
Patent research, strategy advice and drafting of patent applications for AI and computer vision inventions, so start-ups can protect what they build.
Custom advertising design
Professionally designed, custom advertisements that help small businesses attract more customers, online and in their local community.
Ready to deploy
Systems we've already built — adapted to your cameras
Start from something that already works. Each one comes with annotated video, a data file and a dashboard, and we tune it on your own footage.
PPE compliance alerts
Checks hard hats and hi-vis vests on the cameras you already have. Sends a photo alert with the zone and time only when a violation is confirmed, plus an end-of-shift compliance report.
Ask for a demo →Queue & crowd analytics
Head count, queue length, walking speed, people per minute, estimated wait, dwell time and a density heat map, from one fixed camera.
Ask for a demo →Road traffic queuing
Watches a defined stretch of road or highway and detects when traffic starts to back up: where the queue starts and ends, how long it is, and how fast it moves. Alerts when a queue forms, for work zones, incidents and traffic operations.
Ask for a demo →Lane-closure & merge warning
Finds the cones, draws the merge line through them, shades the closed lane and shows the merge direction and distance to the taper in metres. Works from shape alone, even on black-and-white video.
Ask for a demo →Fire, smoke & vehicles in one pass
Several detectors run on every frame and merge into one result: cars, trucks, fire and smoke. Returns annotated video plus a frame-by-frame data file you can search.
Ask for a demo →Corrosion inspection from drone images
Finds rust on large assets from drone imagery, grades how severe it is and measures the percent of surface affected, so repairs can be planned.
Ask for a demo →Trailer fill measurement
A camera at the dock door measures how full each outgoing trailer is: fill percentage, wasted cubic feet and the cost of shipping air per load.
Ask for a demo →Beehive traffic counting
Counts bees going in and out of the hive, spots pollen loads and tracks the drone-to-worker ratio, to show colony health for commercial pollination.
Ask for a demo →Not sure computer vision can solve your problem?
That's exactly what the free 15 minutes are for. Bring a photo, a short video clip or just a description — we'll tell you honestly whether it's worth building, and what a sensible first step would be.
How we work
What happens after the free 15-minute call
No surprises: here is every step from the first call to a working system. You decide at each step whether to go on.
- 1
Free 15-minute call
You tell us what you need to detect or measure. We tell you honestly whether computer vision can do it, and which approach we'd use.
- 2
NDA and a data sample
We sign your NDA. You share a few images or a short clip — or we come on site in San Diego County and collect a sample with our own sensors.
- 3
Short written proposal
Scope, price and timeline for the first step: a fixed-scope feasibility study. You only commit to that step.
- 4
Work in three phases
Feasibility, prototype, then production. Each phase has a clear end point, and you can stop after any of them.
The three phases
Most clients start with a feasibility study, so nobody commits to a build that the data can't support.
Feasibility study
Typically 2–3 weeks
- We review your data, or collect a sample with you
- Success metrics defined in your terms
- Model approach and risks, in writing
- Go / no-go recommendation
Prototype
Typically 6–10 weeks
- A working model on your own footage
- Measured accuracy against the metrics we agreed
- Demo app or dashboard you can show internally
- Weekly check-ins, code handed over
Production & support
Scoped per project
- Deployment to edge, server or cloud
- Integration with your cameras and systems
- Retraining as conditions change
- Ongoing support if you want it
Every phase follows the same loop — get data, clean and prepare it, train, test, improve — and you see results on your own footage before anything goes into production.
Sensors
The sensors most AI shops turn down
Ordinary photos are easy. We work across the spectrum — and we own and operate the equipment. Tap a sensor to see what we do with it.
Who you'll work with
You work directly with the people who build it
Iryna Dzieciuch — Founder & Principal, Computer Vision & AI
Iryna founded IPD Scientific in 2022 and leads every project personally. She has 10+ years of experience building AI vision systems in Python, PyTorch, TensorFlow and MATLAB, across electro-optical, multispectral, thermal, sonar, LiDAR, microscopy and medical imagery — from first feasibility study through deployment, including real-time edge deployment with CUDA, TensorRT and ONNX.
Before IPD Scientific she was a Principal Investigator at the U.S. Navy's NAVWAR and a Senior Data Scientist in 3D imaging at Decision Sciences. She holds 2 issued and 5 filed U.S. patents, has published peer-reviewed research (Springer) on deep learning for sonar imagery, and mentors as a SCORE advisor and science fair judge.
- Leads the technical work, not just the sales call
- Backed by a network of 20+ scientists, engineers and technicians
- 2 issued and 5 filed U.S. patents
- Speaker and mentor: SCORE business advisor and science fair judge. Invite Iryna to speak on computer vision and AI.
Connect with Iryna on LinkedIn →
Pavlo Molchanov, Ph.D.
Senior Consultant — Optics, Electronics & Radar
20+ years in high-speed electronics, electro-optics and RF/microwave design. 10 U.S. patents and 200+ publications; author of Seeing Invisible: Advanced Antenna Arrays. He leads the antenna and radar inventions available for license.
Pavlo has been Principal Investigator on U.S. Navy, Air Force and DoD research programs as a Senior Scientist at AETHER, AMPAC and Compass Systems, and developed new passive-radar, tracking and imaging concepts as Principal Scientist at Planar Monolithics Industries. Earlier he was Head of Department and Vice-Dean at Vinnitsa National Technical University.
Allison Dzieciuch
Marketing & Operations Assistant
Allison supports IPD Scientific's marketing, client communication and day-to-day operations. She studies business at Santa Clara University's Leavey School of Business (class of 2030), where she is also an NCAA student-athlete.
Credentials
StatusWoman-owned small business
Founded2022, San Diego
UEIV1CDSXDEZ6E8
CA contractBID0102538
Earlier work, before IPD Scientific
The experience our team brings to every project.
Corrosion, mine and fake-image detection
As Principal Investigator, Iryna built deep neural networks that found corrosion, underwater mines in sonar images, and fake images, with up to 98% detection accuracy. Her corrosion detection method is now used by the Navy with drones and 3D models.
3D imaging that sees inside vehicles
Improved muon tomography imaging, which uses natural cosmic-ray particles to see inside dense objects. Built exact-size 3D models of vehicles and combined them with LiDAR and photogrammetry in field tests.
Multispectral field data collection
Set up electro-optical multispectral cameras in the field, collected data, and wrote automated image processing in MATLAB: enhancement, segmentation and feature extraction.
Data modeling and forecasting
Modeled data in Systems Tool Kit and built time-series and autoregressive forecasting models in R and Python.
Radar and antenna research for defense
Led U.S. Navy, Air Force and DoD research projects on antenna arrays, passive radar and imaging, resulting in 10 U.S. patents and 200+ publications.
Lab notes
From the lab
What we've learned building these systems, with real numbers.
Clients
What clients say
“I hired IPD to help with my AI Vision System, StrikeSure. In just six short weeks, I am now testing my prototype and building a custom Computer Vision System Algorithm.”
“Working with Iryna and the IPD Scientific team was an exceptional experience… Their expertise in AI-based image processing and real-time detection helped transform our early concept into a working prototype.”
“Iryna's background in computer vision for cameras let her effortlessly understand our challenges and what it would take to address them with AI.”
“IPD is not only a leading-edge firm in the field of machine learning and advanced imaging, they are a pleasure to work with.”
Licensing
Patents available for license
Four pending U.S. patents in multi-beam antenna arrays, passive radar and monopulse SAR — open to licensing and partnership.
Questions
Questions buyers usually ask
How much does a computer vision project cost?
Every project starts with a small, fixed-scope feasibility study, so you know what you are paying for before anything bigger. After the free 15-minute call we give you a price for that first step. Larger prototypes and production systems are quoted phase by phase.
How long does it take?
A feasibility study typically takes 2–3 weeks. A working prototype on your own footage typically takes 6–10 weeks. Production deployment is scoped per project.
Do I need to buy new cameras?
Often not. We start with the cameras and footage you already have. If a different sensor would work better, we own and operate thermal, SWIR, multispectral, LiDAR and other equipment, so we can test it before you buy anything.
Is my data kept confidential?
Yes. We are happy to sign your NDA before you share any images, video or other data.
Do you only work with San Diego companies?
No. We are based in University City, San Diego, and work with clients across the United States. For projects in San Diego County we can also come on site to collect data with our own equipment.
Can you help with an SBIR or DoD proposal?
Yes. We help companies choose sensors, define technical requirements and write the AI and computer vision parts of SBIR and DoD proposals.
What happens on the free call?
You tell us what you need to detect or measure, and you can send a photo or a short clip beforehand. We tell you honestly whether computer vision can do it, which sensor and approach we would use, and what a sensible first step would be.
Based in San Diego
A computer vision company in San Diego
IPD Scientific is based in University City (UTC), near UC San Diego, Sorrento Valley and Torrey Pines. We work with San Diego defense, biotech, medical device, manufacturing and drone companies, and with clients across the United States.
Because we own and operate our own sensors, we can come to your site anywhere in San Diego County, from La Jolla and Carlsbad to Point Loma and Chula Vista, to collect thermal, SWIR, multispectral, LiDAR or drone data for your project.
IPD Scientific, LLC
8910 University Center Lane, Suite 400
San Diego, CA 92122
858-800-2430 · info@ipdscientific.com
Get started
Book a free 15-minute consultation
Pick a time that suits you — no forms, no obligation. We'll talk through your problem and tell you whether computer vision is the right tool for it.
Choose a time
You'll get a confirmation and a calendar invite by email.
What happens on the call
- You tell us the problem. What you need to detect or measure, and what camera or data you have. A sample photo or clip helps.
- We tell you honestly if it can be done. Which sensor and approach we'd use, and the main risks.
- You leave with a clear next step. Usually a small, fixed-scope feasibility study, so you know the cost up front.
Tell us about the project
A few details help us come to the call prepared.