IA and Computer Vision

AI and computer vision: automating complex image and video processing tasks for maintenance, security, health, commerce, the environment, etc.
AI et computer vision

Leverage your visual heritage, exploit every pixel

Every year since 2012, new and more powerful computer vision models have emerged, based on convolutional neural networks or transformers. However, many of these architectures have been designed for their performance gain alone, without considering the operational context or industrial constraints.

We add value here by integrating business knowledge.

AI and Computer vision according to Scalian

Our approach

The importance of being up to date

The AI technology sector, and computer vision in particular, is evolving rapidly. As a Centre of Excellence, we seek to ensure the most effective methods for the specific use cases we encounter.

To achieve this:

  • We carry out technical monitoring and take part in R&D activities within the Group in partnership with Scalian’s Innovation Lab, or through cooperation with academia (in particular with the ANITI institute and the DEEL project). We publicise the results of these partnerships through papers, seminars and conferences throughout the year.
  • Our proposed projects include a component on the state of the art, as early as possible. We do this when the needs and constraints have been identified and the data explored, in order to identify the most recent solutions that are in line with the precise constraints of the use case.
  • In our proofs of concept, we challenge our existing knowledge and architectures, and tailor our proposed architectures to the state of the art.
  • Our team is mainly made up of PhDs, ensuring our ability to interact with the state of the art in research.

Computer vision use cases

Artificial vision coupled with AI can be used to address a whole series of complex functions, for which the data consulting team has developed solid expertise:

Classification: sorting images into different categories. Automatic sorting of products, drugs, vegetables according to images of these items. For example, the data consulting team has had great success with the following:

  • Sorting the road type on routes travelled by a vehicle, to determine its use profile for predictive maintenance purposes.
  • Detecting images of undesirable products in an e-commerce marketplace, to avoid users selling weapons and drugs.
  • Differentiating between product images with white backgrounds and product images where white borders have been added, to automatically harmonise the graphics.
  • Sorting images of military vehicles on SAR images, with a model trained using images generated by a Scalian simulator (cf page CEN Simulation).

Detection of features of interest in images. We carried out the following detection projects:

  • Flying drones carrying systems that automatically detect humans, animals and vehicles, to safeguard the release of sensors. Simultaneous detection from oblique and nadir images in visible and infrared.
  • Automatic, real-time detection of passing trains to allow operators to rapidly analyse sequences of trains in surveillance videos.
  • Detection of faults on solar panels from thermal images.

Tracking: detecting and monitoring one or more objects of interest in videos. For example, in this context we focused on:

  • Detection of illegal waste dumping in surveillance videos.
  • Automatic traffic analysis: congestion and average speed, via vehicle tracking in drone videos.

Segmentation: defining specific areas of pixels of interest in images. This is used, for example, to determine different types of soil, environments, or housing in satellite images; or to focus on people or objects of interest in photos and videos. As such, our team worked on the following application:

  • Determination of the number of trees felled and stacked in a forestry context. It then became possible to calculate the volume of timber harvested by focusing on the logs in the images.

Our method 

Auditing and assessing use cases

Before starting to integrate AI, it is essential to assess its relevance. Just because deep-learning methods offer leading performance in many image-processing cases does not mean they should systematically be applied to all use cases. Moreover, even in cases where deep learning proves to be the best choice, it is important to accept that not all of its applications will necessarily provide the same value, or be accepted in the same way by their users.

After six years in existence and applications in a wide range of fields, we can help you identify, assess and clarify your needs. Following our audit, you will be given a summary enabling you to assess the feasibility, viability and acceptability of AI applications for your use cases, as well as our recommendations on the steps to be taken, and on the relevant and irrelevant approaches. If AI is not the best solution for your needs and constraints, we will explain why!

Integrating your constraints

We have the know-how and experience to identify and integrate your constraints – whether functional, hardware or performance – into the solutions we implement.

***Real time or embedded, we’ve been there already!***

Scalian resolves to take account of the specific industrial context before embarking on the mindless pursuit of performance, whether upstream, through the choice of approach and definition of the architecture, or downstream, through the reporting of AI output and the assessment of its performance. If, for example, the false-positive rate is critical to your business application, the performance of the models we offer should reflect this.

Towards a more transparent AI

One of the risks of AI-based solutions is the “black box” effect, where you have no visibility of how our proposed solution works, its limitations, and therefore how much you can trust it.

We have several ways of minimising this effect:

  • Our project methodology, based on agile operation, offers regular ritualised communication with your experts, in which we set out in simple terms what we have done and what we are going to do, throughout the project. At the end of the project, your experts therefore have a better idea of the solution, as they have been able to follow all the steps and choices leading to its realisation. These exchanges also enable you to ensure that the solution incorporates your constraints.
  • Wherever possible, at the start of the project, we suggest one or more acculturation meetings, during which we can explain in simple terms the AI and the methods we intend to use for the project.
  • Our team is closely involved in fundamental research on topics such as the explainability of AI models, and we are working to integrate methods on explainability or on assessing the robustness of our models.

The guarantees of a bigger service company

Innovation should never be synonymous with low-quality code in the delivered solution. To ensure this, our team takes advantage of Scalian’s Digital Factory: a group of over 40 experts producing software solutions. We are therefore able to pay particular attention to the quality of our code and environments, and to the adaptability of what we produce, in order to facilitate any upgrades or retraining of the AI we offer.

It also means we have the resources to be able to offer to train complex AI on our systems, locally or in the Cloud. Our multidisciplinary team directly includes all the various professions needed for every step of an AI project. This allows us to ensure that projects run smoothly, involving all the necessary players from the outset.

Helping you develop skills in AI

Whether you want to set up in-house computer vision projects, become acculturated, train your experts, or allay any fears and reticence towards the use of AI, Scalian offers to help with:

  • We have a small team of experienced trainers, and offer customised or off-the-shelf training in areas such as Python, machine learning, deep learning, natural language processing, etc.

We have worked both for large private groups (Orange, Renault Lab, etc.) and in academic environments (ENAC, Ynov, ENSEIRB-ENSC, etc.).

  • Consulting: Meet and talk with one of our experts to secure the launch of your upcoming projects, bolster the expertise of your own teams, or rescue a project in difficulty.
  • Expertise: On a fixed-price basis or working closely with your teams, we offer our extensive know-how on the design and development of innovative solutions.