OPT DeepVision3 solves the problem of AI implementation

Mondo Education Updated on 2024-01-30

OPT DeepVision3 integrates the visual basic large model, which not only improves the robustness of the model, but also greatly shortens the cycle from training verification to deployment, and makes the annotation interaction and various functional tasks more convenient, which solves the pain points of deep learning Xi in industrial production.

Efficient

AI models are faster and lighter to train

How to reduce data dependence, labor costs, application thresholds, and shorten the total cycle has always been the primary problem hindering the widespread implementation of deep learning Xi.

In order to overcome these challenges, DeepVision3 continuously optimizes the underlying logic algorithms, and realizes key technological innovations in incremental learning Xi, small-sample Xi learning, and model lightweighting, which greatly reduces the time cost of data collection, model training, and migration.

In the face of the vision scheme with fewer defective samples, DeepVision3 reduces the amount of data by 90% by using small-sample strategies such as data augmentation and algorithm enhancement, from a few hundred in the past to more than a dozen, or even a few to complete AI model training. A large number of high-quality training images are generated based on the deep image generation network, and the generation speed is increased by more than 3 times.

Image generation model.

Under the premise that the model performance is almost unchanged, the model training can be completed in 30 minutes for 4K scale dataIn order to better meet the application requirements of industrial scenarios, DeepVision3 can realize incremental training of new requirements in just a few minutes.

In addition, DeepVision3 not only reduces the computing power requirements and inference time through the model lightweight strategy, but more importantly, makes the model detection accuracy higher.

Schematic diagram of deep learning Xi model training.

With a CPU, it takes about 60 ms to detect 20 million pixels of key targets. Compared with conventional algorithms, the inference speed of detection and classification tasks is increased by more than 20 times.

Flexible

Integrate the visual foundation model to fit the factory model

While making the software more efficient, OPT also uses technologies such as transfer Xi and domain adaptation to ensure that the trained model is more flexible, integrating generalization, generality, and flexibility.

In the face of quality inspection with similar processes, DeepVision3 can realize one-click changeover based on one-click migration technology or adaptive fine-tuning, and the training period can be shortened to several hours, which solves the problem of poor model generalization caused by large differences in defect morphology and frequent product model changes.

Schematic diagram of one-click migration technology.

For the 3C and lithium battery industries, OPT has also developed a general detection model, and the key process defect detection can be used out of the boxAt the same time, the Zhixin large model will be launched soon, which can locate and detect key objects with a new detection method without model training, further accelerating the wide implementation of AI detection in more industries.

Not only that, deepvision3 also supports functions such as global management, multi-person collaboration, multi-process analysis, and multi-machine collaboration, which is highly suitable for the existing factory generation mode requirements.

Easy to use

AI is feature-rich and can be deployed with one click

DeepVision3 includes a variety of task types such as semantic segmentation, character recognition, object detection, and image classification, which requires no programming and is highly easy to use, which greatly reduces the cost of software learning Xi.

DeepVision3 is equipped with a number of intelligent auxiliary annotation tools. For character recognition tasks, DeepVision3 has built-in general OCR and centralized inspection functions to achieve semi-automatic annotation of characters, which can recognize characters in any direction or multiple lines of text with angles, and users only need to check the results.

Character recognition schematic.

At the same time, for the most time-consuming semantic segmentation annotation task, semantic segmentation AI tools, deep learning Xi automatic annotation, traditional algorithm automatic annotation, contour extraction, etc. are integrated. Among them, the semantic segmentation AI tool can automatically generate high-precision and accurate pixel-level object annotation based on the user's point of interest, target box, and mask information with just a click of the mouse or pull a box.

Schematic of semantic segmentation.

In addition, deepvision3 also supports multi-label reuse, annotation quality control and other functions. In the process of model training, tools such as hyperparameter setting tips, process visualization, and evaluation result traceability are providedIt can also be deployed to smart3 software with one click.

For more details about deepvision3 technology and applications, please pay attention to *** and the official website.

Related Pages