Transwarp is committed to building enterprise-level big data infrastructure software, providing enterprises with infrastructure software and supporting around the whole data lifecycle to build a data world of the future.
Sophon AutoCV is a computer vision model development software under Sophon, providing users with one-stop CV modeling services, helping users to reduce modeling costs and improve development efficiency.Sophon AutoCV takes raw data as the starting node, analysis model as the tool, and data value as the business end point. By building a "data-model-feedback" closed loop, the CV model can be efficiently and continuously iterated and eventually deployed at scale.
What can Sophon AutoCV offer?
Support you to customize file storage paths and preview or export sample sets. And support you to add, delete, modify, and query a sample set.
Support for common image annotation tasks such as image classification, target detection and image segmentation. Support user-defined labels and preset rich shortcut keys, which facilitates annotation. Provide intelligent model pre-annotation function, which greatly improves the annotation efficiency.
Support multi-person collaborative annotation tasks; Provide annotation task distribution, annotation progress management, annotation content audit and other functional modules, which is convenient for team management, ensuring annotation quality and clarifying the responsible person of annotation.
Support the selection of training data across annotation sets to generate datasets, as well as dataset preview, label combination and mapping, test sets customization and other functions.
Guided Model Training
Provide pre-built algorithm frameworks and pre-trained models for both classification and target detection tasks, as well as a guided model training service to help users quickly deploy the trained models as a service.
Support for model testing and evaluation via default test sets/custom test sets, and output of multiple evaluation metrics. And support multiple versions of model performance comparison and visualisation of model evaluation results to help users find the best model quickly.
Be able to connect the training process to the model service, so that the trained models can be managed in the model warehouse with one click. Support model encryption and multi-version management, and support model one-click deployment to the server for landing use.
Performance Monitoring and Model Iteration
Support real-time monitoring, exception alert and data return of deployed models to facilitate iterative model training, At the same time, support automatic and intelligent iterative updating of models.
Flow Diagram of Model and Training Data in Sophon AutoCV
Why should you choose Sophon AutoCV?
One-stop Model Development to Reduce Modelling Costs
Sophon AutoCV covers the entire process from sample management, data annotation and model training, meeting the needs of users in the modelling process without the need for cross-platform operations. The process is compact and cost effective, while being highly adaptable to fragmented scenarios.
Build Models with Low Code to Improve Development Efficiency
Sophon AutoCV supports a full visual guided modelling process. Industry-leading algorithms and high-precision pre-trained models are pre-built for different CV scenario tasks and ready to use. Highly accurate CV models can be trained without programming.
Train Models with High Performance to Support Rapid Modeling and Reasoning
Sophon AutoCV uses GPU to accelerate model training and can accelerate reasoning based on heterogeneous hardware. It supports many domestic and overseas mainstream acceleration cards. Model training and reasoning can be further accelerated by optimizing GPU utilization. At the same time, TensorRT framework is used to accelerate model inference to achieve high performance model reasoning.
Launch the Model with One Click for Rapid Application Deployment
Sophon AutoCV has been integrated with Sophon Edge which can put the trained models into model warehouse on the cloud node with one click. After being put into Edge, the models on cloud can be delivered to Edge Nodes and deployed.
Enterprise-Level Modeling Platform to Ensure Data Security
In terms of data security, Sophon AutoCV sets up sample watermarks to prevent data leakage by annotators, records audit logs to prevent malicious manipulation, and encrypts generated model files to ensure model security. Meanwhile, the platform ensures user data security by establishing user authority management system and data isolation.
Model Training in a Research Institution
Teaching at Polytechnic
Using in Transwarp Algorithm Team
○ A research institute wanted to identify the types of multiple creatures captured in a video. The manual frame-by-frame recognition of video content is labour-intensive and the workload is huge
○ The factors, such as seasonal changes, lighting changes, etc, can make the same model unsuitable for all scenes, so the model needs to be updated regularly
○ Through data management, data annotation to model training and other functions in Sophon AutoCV, the biodiversity recognition related models have been trained successfully.
○ During the deployment process, the Edge platform supports data monitoring and retransmission to enable model optimisation and update
Biodiversity models trained on Sophon AutoCV platform have been tested well, with an average accuracy above 96%
During the deployment process, data is monitored and returned through the Edge platform to support the optimization of the models, which ultimately enables the models to perform well in different scenarios.