Getting started with CVOps
What are CVOps?
CVOps stands for Computer Vision Operations. These kinds of operations lie at the intersection of Machine Learning Operations (i.e., MLOps) and Computer Vision (CV) processes. When you apply MLOps to CV, you are performing CVOps!
How can CVOps help my business?
CVOps encompass the tools and processes necessary to develop, deploy, and maintain CV models on a large scale.
Using Picsellia, you can boost the efficiency of your team by allowing them to work together on a single platform.
The reliability and the traceability of your CV projects are ensured by the fact that each step is orchestrated by your team on Picsellia.
The success rate of your CV projects will increase and the development time will drop. Picsellia is freeing data scientists from low-value tasks!
By setting up automated data pipelines, the delivery time of improved model versions will drastically decay.
How can I know which predictions are relevant to review and add to the feedback loop?
Each prediction sent to Picsellia goes through our monitoring system which provides you with valuable insights. You will be able to decide whether it is relevant to review an image and use your reviewed data as new training data!
What is Picsellia?
Picsellia is an MLOps platform dedicated exclusively to Computer Vision (CV) projects. Our mission is to help technical teams to upgrade their CV models into CV pipelines. Picsellia is your ally at each step of your CV journey, from the moment you start gathering your data in one single place to the deployment & monitoring of your model in production.
Can I train my own models?
Yes, Picsellia offers the possibility to integrate your own models. If you use standard architectures such as TensorFlow Object Detection API or YOLOv5, v7, v8 you will be able to upload your own weights and checkpoints. If not, you can integrate your own custom models and training script through Docker.
Can I run training on my own GPU?
Yes you can! We generate docker commands with the proper flag for you to run your training on your own GPU clusters.
Can I deploy Picsellia on premises?
Yes, Picsellia supports on-premise deployment. You will require a Kubernetes cluster.
Are there any ready-to-use models available?
We do provide models ready to use for standard COCO classes, but we are not a recognition API provider.
What is unique about you?
Picsellia was created specifically to handle unstructured data such as images. This kind of data requires a dedicated platform to be indexed, visualized, and annotated. The metrics Picsellia provides to assess dataset quality, experiment success, and model performance are specific to CV use-cases.
Will my data be safe?
Picsellia will soon get the ISO27001 and SOC2 certificates.
You can use Picsellia to read the data that are stored on your own infrastructure. If you prefer, you can also use the Picsellia infrastructure to store your data. Our infrastructure is located in France and operated by OVH Cloud.
What about the ownership of data and models?
Picsellia is a platform that allows you to create your CV models, therefore, you remain the owner of the data you use and the models you create on Picsellia at all times.
Every single data generated through Picsellia (annotations, models, predictions, etc.) remains with our customer’s IP.
In order to ensure data security, we also offer our customers the possibility to connect their own storage.
Scaling with you
Can I customize the training of my model?
You can modify all the hyperparameters of our pre-packaged models. If this is not enough, you can package your own models and modify any parameters you want through our Python SDK.
Can I customize my experiment tracking dashboard?
You are able to log everything you want through the Python SDK. If you use our pre-packaged models, everything will be already logged and it will be up to you to filter out what you do not need. You can use all the following log types to personalize your experiment tracking dashboard: line, bar, chart, images, confusion matrix!
Who’s currently using Picsellia?
Computer Vision can be applied to many domains. Picsellia’s main customers are large-scale industrial firms and Computer Vision provider companies. Our clients use Picsellia to build their CV models. The sectors of our main customers are manufacturing, energy, retail, security, smart city, farming, and sports analytics.
I’d like to know more about you!
Picsellia is VC-backed by renowned tech investors across Europe and the US.
It is a 3-year-old company founded by two engineers from Toulouse. The team currently has more than 15 people located around Europe. Our headquarters are in Toulouse and we have offices in Paris. Know more about us!
What is Picsellia’s business model?
Our business model is based on annual licenses. The license price depends on the service level and the usage of the platform. Platform usage is measured in DPU and MPU.
What is a DPU?
DPU stands for Data Processed Unit. It is Picsellia’s unit of measurement for the amount of data processed on Picsellia (indexing, embedding creation, monitoring of outliers and drift).
1 DPU = 1 GB Data Processed.
What is a MPU?
MPU stands for Model Processed Unit. It measures the amount of running time of your models (for pre-annotation, training, predictions, etc.).
1 MPU = 1h of models running.
Can I try Picsellia before I make my decision?
Of course! We value the satisfaction of our customers and we want to make sure Picsellia becomes an added value to your company. You can claim your free trial at any time.