ViMo Deeplearning
Product Description
ViMo Deeplearning is a desktop industrial vision deep learning training software developed by SmartMore. Leveraging the cutting-edge computer vision algorithms from ViMo Cloud, the product focuses on refining and optimizing smart annotation, automatic algorithm training, rapid model tuning, and efficient SDK inference. Additionally, it is user-friendly and requires zero programming skills, allowing users to accomplish core tasks in smart manufacturing production lines, such as material classification, defect detection, object localization, and character recognition.
As a desktop software, ViMo Deeplearning complements ViMo Cloud by providing offline model training capabilities. This achievement enables the ViMo product series to offer comprehensive coverage of capabilities in the industrial vision field, empowering more enterprises with opportunities for visual intelligence upgrades and facilitating the practical implementation of intelligent manufacturing upgrades.
Product Features
Advanced Smart Annotation, Boosting Operational Efficiency:
ViMo Deeplearning offers a range of image annotation features, including AI-powered annotation, non-learning regions and label soft/hard merging. With one-click AI-powered annotation, average efficiency can be improved by over 70% while maintaining the same annotation accuracy.
On-site Data Processing with Enterprise Privacy Protection:
By leveraging the capabilities of edge computing, ViMo Deeplearning enables on-site processing of production line data and model construction without the need to copy data off-site. This approach effectively safeguards enterprise production privacy, eliminating concerns about the leakage of confidential information such as materials, equipment, and production line plans.
Model SDK Export for Stable and Flexible Development:
ViMo Deeplearning supports one-click deployment of models for real-time inference and operates seamlessly in multiple development languages and system environments, effectively accommodating various hardware types. By leveraging techniques such as model distillation and pruning, it creates compact yet powerful models that stably run in different scenarios, including intelligent vision controllers, smart cameras, and industrial computers.
In-depth Data Analysis for Intelligent Decision-making:
The software provides visual analysis of model inference data and interactive post-processing capabilities. Through visualized graphs, users can gain insights into global information regarding false positives, false negatives, and defect statistics. Real-time adjustments to various parameters can be made based on inference results, allowing for quick alignment with false positive rates, false negative rates, and other indicators. This enables flexible project delivery.
Edge + Cloud, Overcoming Production Line Constraints:
In offline environments, ViMo Deeplearning complements ViMo Cloud by providing offline model training capabilities. This comprehensive coverage in the industrial vision field ensures data format, model format, and solution format compatibility between the two products. ViMo Deeplearning projects can seamlessly integrate with ViMo Cloud, allowing users to enjoy the convenience of cloud services effortlessly. It provides more enterprises with opportunities for visual intelligence upgrades, facilitating the practical implementation of intelligent manufacturing upgrades.
Application Cases
Precision Grinding Surface Defect Detection for Steering Joints at a Global Top 500 Japanese Automotive Parts Manufacturer
The automotive steering knuckle forging process is complex and prone to forging defects, resulting in defective or even scrap products, causing significant financial losses for the company. If these products enter the market, they can directly impact the occurrence rate of traffic accidents. Therefore, the manufacturer needs to develop the capability to quickly identify and eliminate forging defects on the automated production line.
Based on the core algorithm capabilities of SMore ViMo Intelligent Industrial Platform, SmartMore has provided a comprehensive solution for precision grinding surface inspection of steering knuckles to the customer. With the Vi-Lab operation management system, SmartMore enables the customer to quickly build a machine vision system and deploy it rapidly. The hardware components include light sources, cameras, etc. This solution achieves an accuracy rate of over 95%, resulting in a cost savings of 60% for the customer. SmartMore has successfully implemented the first automated intelligent quality inspection project for this manufacturer, serving as a successful pilot project.
USB Defect Detection for a Well-known Consumer Electronics Manufacturer
The project requires a three-class classification (OK/NG/NG-2) of scratches and dirt on the USB interface. Previously, the customer's quality inspection was mainly carried out through manual visual inspection, which resulted in high costs and low detection efficiency. Additionally, the tested products have various types of defects, requiring precise defect classification based on their characteristics during inspection. Traditional computer vision algorithms cannot establish standard detection templates for some defects, necessitating the use of high-precision deep learning algorithms for detection.
Based on the high-precision algorithm capability of SMore ViMo, SMIoT can perform multiple defect detections on products with an accuracy and recall rate of up to 98.9%. The false negative rate is below 2%, enabling the realization of fully automated production lines. This solution effectively reduces production inspection costs, improves product yield, and ensures operational efficiency.
OCR Smartwatches from a Top Global Smartphone Manufacturer
The project requires the recognition of characters engraved on metal components of smartwatches.
There are three different fonts for the characters, and the strap can come in black, white, or orange
colors. The imaging conditions are complex, including various forms such as blurriness, lighting
variations, and tilting, presenting significant challenges for recognition.
Based on SMore ViMo's industrial optical character recognition algorithm, SMore has developed a
sophisticated preprocessing approach to enhance the algorithm's robustness to environmental
variations. This approach ensures a high character recognition accuracy of up to 99.9% and enables
compatibility with various product configurations, As a result SMore has successfully implemented fully
automated production on six production lines for our customers.
