Developed by CIM AtlantiqueThe industrial vision software that brings together every AI tool.
Deeper brings together in a single platform the best artificial-intelligence and vision algorithms — anomaly detection, OCR, measurement, counting, vision-language models — and assembles them without a single line of code.
score and heat map per part
Deeper succeeds Deep Capture. The supervised deep learning proven on our customers' lines remains at the heart of the tool, now surrounded by a complete toolbox and a no-code editor.
The right algorithm for every inspection.
No single algorithm can do everything. A surface defect, a dimension, a marking or a count each call for different tools: Deeper brings them together and lets you combine them.
Every approach, in one place
Deep learning, anomaly detection, rule-based "classic" vision, OCR, polarization, VLM: you pick the technique that fits the defect, not the other way around.
An application is built, not coded
Each inspection is a visual pipeline: you connect blocks, test on your own images and tune the parameters live.
Built for the production line
GPU computing to keep up with the line rate, multi-camera, a customizable operator screen and direct dialogue with the PLC.
More than 80 blocks, ten tool families.
From latest-generation AI models to proven vision tools, each block slots into a pipeline and combines with the others.
Anomaly detection
Learns what a good part looks like and flags any deviation, even a never-before-seen defect, with a score and a heat map.
Detection & classification
Locates, classifies and sorts objects and defects in real time, including from a simple text description of the object you are looking for.
Segmentation
Precisely outlines each part or area to measure a surface, isolate a region or straighten a label.
OCR & codes
Reads back dates, batch numbers and markings, even irregular inkjet or slanted text. Reads barcodes, QR and Datamatrix codes.
Measurement & geometry
Sub-pixel edge measurement, line and circle fitting, distances, angles, part alignment and pattern search.
Counting
Counts objects from a few examples or a description, including stacked products.
Color
Checks the hue of an area against a reference and analyzes color distribution.
Polarization & reprocessing
Removes reflections, reveals micro-defects and prepares the image: contrast, filters, straightening, cropping.
VLM — vision by prompt
Vision-language models that read, describe and answer a question about the image. Ideal for getting an inspection started quickly.
Decision & communication
Thresholds, AND/OR logic, final OK/NOK verdict, then sending the result to the PLC or the information system.
The right tool, not a single tool. On one and the same part, Deeper can chain a straightening step, an anomaly detection, a batch-number read and a measurement, then return a single verdict to the machine.
From image to verdict, everything happens in Deeper.
Six modules cover the full cycle of an inspection: describe the object with a prompt, run production, build the recipe without coding, organize and annotate images, train the models. Deeper connects to your industrial cameras and sends its verdict straight to the PLC.
The software, module by module
Hover over a tile to see it in action.The tiles animate as they appear on screen.
VLM_PROMPT
✦ VLM promptDescribe the object to detect
Prompt "syringe" · analyzing…
✓ Detected without training
01 Vision by prompt
Describe what you're looking for, without training a model
With vision-language models (VLM), you simply describe the object in a few words: "syringe", "cap", "label". Deeper detects and outlines every occurrence in the image, with no annotation or training. Ideal for starting an inspection in a few minutes or pre-annotating a dataset.
02 Operate
Run production and inspect in real time
The production screen. An acquisition is triggered, by the operator or the PLC, and Deeper runs the inspection: detections, verdict and processing time for each step appear immediately. Live mode on camera, or Simulation on recorded images.
03 Studio
Build the inspection, without coding
You drag blocks from the palette (camera, preprocessing, AI, OCR, measurement, counting, decision, PLC), connect them, then compile and activate. Each recipe is saved as a reusable preset, and the Layouts tab composes the operator screen.
04 Data
Organize and trace your images
Image collections, datasets, versioned snapshots and training runs: the whole chain can be read at a glance, from the raw image to the model, with the history of every change.
05 Annotation
Annotate fast, with AI assistance
Oriented boxes, fill, Smart click: one click on the object and the assistant draws the annotation. It also finds similar images or pre-annotates with your own model, and tracks progress and the train / validation / test split.
06 Learning
Train your own models
You pick a dataset version, set the hyperparameters and launch training: YOLO (detection, oriented boxes, segmentation) or anomaly detection. Each run is logged, evaluated and comparable before going into production.
See Deeper on your parts
Send us a few images of your products: we build the inspection in Deeper and show you the result.
Test Deeper on my parts → Request a demo
One inspection = one visual pipeline.
In Deeper's editor, you drag the blocks onto the canvas and connect them: camera, preprocessing, AI, measurement, decision, PLC output.
Six steps, one single platform.
Acquisition, training, fine-tuning and deployment all happen in the same software: no tools to stitch together, no data to move around.
Acquire the images
Live from the line cameras, or remotely from a phone or tablet for the first trials. Images are organized into collections.
Annotate
Defects and reference parts are annotated directly in the tool, with data augmentation to enrich the image sets.
Train
Models are trained online, runs are tracked and versions compared: the best model is promoted to production.
Assemble the inspection
The blocks are connected in the visual editor: preprocessing, models, measurements, decision rules.
Validate on real images
The pipeline is replayed on the saved collections to tune thresholds and measure performance before going live.
Deploy
You compose the operator screen (video feed, results, permitted settings) and deploy to the line workstation, connected to the PLC.
Inspections that used to be out of reach.
VLM · locating & countingStart an inspection with no training campaign.
Vision-language models (VLM) connect what they see to words. Here, loose syringes: the model locates each syringe, even when overlapping others, and Deeper counts them, with no prior annotation campaign.
The VLM is not a magic wand: it is one more tool in the box. Depending on the inspection, Deeper combines it with trained models and measurement tools to reach the reliability expected in production.
Learn more about VLMs
What a VLM can do in Deeper
- Locate objects named by their name, and produce a mask per object.
- Count loose parts, even when they overlap.
- Read difficult text and markings: slanted, irregular, on a busy background.
- Sort between several labels: pass / fail, reference A / reference B.
- Answer a question about the image: presence of an element, condition of a product.
Two examples
When to use it
For a feasibility study or a first demonstrator in a few hours, when product references are numerous or change often, and when the object or criterion is easy to describe.
Its limits, and how Deeper makes up for them
A VLM is slower than a dedicated model and less precise on very subtle defects. For high line rates, Deeper switches to a trained model: the images already collected and sorted by the VLM serve as the training base. In every case, the final verdict goes through decision rules and thresholds that you control.
Test a VLM on my images
Polarization · reprocessingReflections are no longer an obstacle.
Glass, lacquer, plastic films, polished metal: glossy surfaces defeat most vision systems. With a polarized camera, Deeper computes the Stokes parameters, removes reflections and brings out dents, scratches and dust.
On lacquered pens, defects smaller than 0.1 mm become visible; on film-wrapped trays, OCR reads back dates and batch numbers printed through the film.
To go further on reflective surfaces, Deeper pairs with Deep Lighting, our dedicated lighting solution.
Technical sheet.
A very wide range of challenges.
Glassware, cosmetics, wood, food, packaging: a few applications in service at our customers' sites.
Inspection of bottle bases
Incisions, seed bubbles and burn marks detected around the entire base rim, distinguishing real defects from mold seams.
Micro-defects on lacquered pens
Image reprocessing removes reflections and highlights dents, dust, scratches and lacquer defects.
Board inspection
Knots, cracks and missing material graded in real time, down to telling "spike" knots from standard knots.
Bare cheeses
Foreign bodies, cracks, missing material, dimensional and rind defects, on products moving along the line.
Chicken drumettes & wings
Hemorrhages, bruises, exposed bones, shape defects: each piece classified without slowing the line.
OCR on plastic films
Polarization to remove film reflections, then AI able to read back irregular inkjet printing and barcodes.
Club sandwiches
Checking that each sandwich is complete: no missing slice or filling.
Reading difficult markings
Engraved or printed characters on curved and glossy surfaces, read back character by character.
Automatic counting
Counting of items in an image, with no rule to program, from just a few examples.
Turnkey, or in your hands.
CIM Atlantique has been designing special machines, robotic cells and vision systems for more than 30 years. Deeper can be used in two ways.
We design and integrate your inspection.
Our engineers master the whole chain, from the feasibility study to on-site commissioning.
- Feasibility study on your images and your parts
- Selection of lighting, optics and camera
- Development of the Deeper application
- Mechanical, automation, ejection and robotic integration
- Commissioning and performance monitoring
Your teams take over.
For integrators, process-engineering departments and quality teams who want to build and evolve their inspections on their own.
- License for the Deeper software
- Remote onboarding training over 2 to 3 days
- Training eligible for funding through your OPCO, via our Qualiopi-certified partner organization
- Support from our engineers on your first applications
Frequently asked questions about Deeper.
A question that isn't here? Call us on +33 2 40 58 72 96.
Does Deeper replace Deep Capture?
Yes. Deeper is the next generation of our AI vision software. It builds on the experience gained with Deep Capture in supervised deep learning and broadens it: anomaly detection, measurement, OCR, counting, polarization, vision-language models, all programmable without code. Do you have a Deep Capture installation? Let's talk about upgrading it.
Do you need to be a vision or AI expert?
No. Inspections are built without code, by assembling blocks in a visual editor. Our engineers can deliver the application turnkey, or train your teams so they can evolve it themselves.
How many images do you need to get started?
It depends on the technique chosen. Anomaly detection learns mainly from good parts, which is valuable when defects are rare. Vision-language models sometimes make it possible to start with no specific training. For supervised models, we define the image campaign with you during the feasibility study.
Which cameras are compatible?
Deeper works in particular with Baumer, Basler and SVS-Vistek industrial cameras, including polarized ones. It can also receive images remotely (phone, tablet) for the fine-tuning phase.
Cloud or on-premise?
Both: on-premise on a PC equipped with a GPU, as close as possible to the line and without relying on the internet, or as SaaS on a web server.
How does Deeper communicate with my PLC?
The inspection result (OK/NOK verdict, measurements, read text) is sent to the PLC or the information system via OPC-UA, Modbus TCP or an HTTP request.
Can we test Deeper on our parts?
Yes. Send us a few images of your parts, both good and defective: we assess feasibility with Deeper and present the results to you, by video call or in our workshops in Grandchamps-des-Fontaines.
A difficult inspection? Let's test it with Deeper.
Reflections, random defects, a wide variety of references: send us a few images and we'll tell you what Deeper can do.
44119 Grandchamps-des-Fontaines
