Introducing Picsellia v3.0

Introducing Picsellia v3.0

Introducing Picsellia v3.0

Our Mission

We’ve created Picsellia for one reason, and one reason only, we strongly believe that Computer Vision will be everywhere shortly. However, the need for Computer Vision is way bigger than the talent pool in AI. 

The only way to do more with limited resources is to optimize the way of working. Looking back, there is only one answer to this challenge, tooling. Efficient tooling helped people build roads and bridges, and bring electricity to the world. Now tooling is helping Companies deliver value with AI to their customers. 

Our mission is to empower Computer Vision Engineers with the proper tools to do more, with the same team. 

We articulated this vision with a cloud and on-premise platform that provides all the tools a Computer Vision team will need to generate value. Everything is covered and designed to allow engineers to gain time and performance from data management to model monitoring. 

As the market and Computer Vision usage evolve, the associated tools need to do the same. We have been discussing Computer Vision needs and usage with our customers daily, and understand that a shift is coming. So we decided to adapt.

Pains in the market and product gap 

The computer vision market is fascinating nowadays. We saw an exponential growth of use cases in enterprises that became more concrete, scaled, and profitable. But as this industry expands, the underlying complexity increases, and acquisition devices generate bigger, heavier images, with more information to exploit. 

However, there’s one downside, it’s data chaos.

This problem existed before, and many companies tried to tackle it. Unfortunately, it got worse. More data and more information, but nothing in the tooling market to help enterprises navigate and understand this chaos at scale is found. For some cases, it’s not a big deal; just feed an LLM with as much data as you can, cross your fingers, and hope it will do the job. Unfortunately, Computer Vision doesn’t work like that (here is a study on VLM) for real-world use cases.

The other effect of the increased data complexity is that the deep learning architectures (driven by the transformer revolution) are getting more and more complex, and harder to use. At Picsellia, we used to say, “Training a model is a commodity nowadays”. And although it was true 2 years ago, it is not the case anymore. One needs to have expensive computing power, distributed computing skills, and a deep understanding of signal processing to navigate the new AI challenges. The rise of embeddings paved the way for multi-model AI models that can aggregate data from different sources, channels, and use cases in one single approach. 

The last thing we saw was that Computer Vision projects are now on a bigger scale. The amount of visual data poured into CNNs or ViT is exponentially bigger than before, creating flow in a lot of tools that are not capable of handling the data load and project scale.

Filling the Gap in the Annotation Market

Welcome to the next big thing in computer vision - Picsellia Platform v3.0! The demand for sophisticated data tools has exploded as the industry evolves at a breakneck pace. Traditional annotation methods just aren't cutting it anymore, creating significant roadblocks for development teams eager to push the boundaries of computer vision technology. Enter Picsellia v3.0, our latest release, packed with innovative features designed to completely transform the annotation landscape.

Annotation Campaign & Video Labeling

One of the highlights of Picsellia v3.0 is our enhanced annotation capabilities, with a special emphasis on video labeling. Given the pivotal role of video data in developing advanced computer vision models, our platform now delivers comprehensive tools for efficient video content management and annotation. This upgrade not only streamlines the annotation workflow but also guarantees that models are trained on accurately labeled data, boosting their effectiveness and dependability in practical applications.

EU Compliant Tool

In the global digital marketplace, adherence to regulatory standards, particularly the EU's stringent requirements, is non-negotiable. Picsellia v3.0 is engineered with these standards in mind, ensuring that your data management practices are secure and privacy-compliant. Additionally, our platform's broad support for various data formats caters to the rich diversity of visual data encountered in computer vision projects. Whether it's standard image files or specialized formats like DICOM and TIFF, Picsellia v3.0 ensures seamless integration and processing, empowering engineers to innovate freely without being hindered by compatibility issues.

Data Projection With Multi-Layer Support

Introducing data projection, a revolutionary feature of Picsellia v3.0 that enables users to overlay annotations or metadata across multiple layers of a single image or video frame. This capability is invaluable for projects requiring intricate analysis and annotation, offering engineers a more nuanced and flexible approach to data visualization. With this feature, teams can gain deeper insights into their data, leading to the development of more accurate and innovative solutions.

Leveraging LLMs With Picsellia GPT and Visual Prompting

The advent of Large Language Models (LLMs) has significantly expanded AI's ability to understand and generate human-like text. Picsellia v3.0 leverages this potential through Picsellia GPT and Visual Prompting, integrating the sophistication of natural language processing to refine the annotation process. Picsellia GPT automates the generation of tags and annotations by comprehending the visual data's context, drastically reducing manual effort and enhancing efficiency. Meanwhile, Visual Prompting offers an intuitive interface for users to guide and fine-tune the annotation process using natural language. These innovations streamline operations and pave the way for new methodologies in training computer vision models.

Want to learn more? 

Check out the interview with our CTO, Pierre-Nicolas, where he dives deeper into the new functionality we released

Get a free copy of our Comprehensive Guide to Video Labeling: Tools, Functionality and Techniques

Start managing your AI data the right way.

Request a demo

Recommended for you:

french language
english language