Marketing Personalization · Real-time Analytics
Lenztech had to process millions of retail shelf images and massive SKU catalogs every day while maintaining very high recognition accuracy and low latency. Complex in-store scenes, frequent model updates, and the need for better small-file handling meant the company needed a faster data pipeline, stronger inference performance, and higher GPU utilization without turning platform operations into a bottleneck.
Reported impact
The Alibaba Cloud redesign helped Lenztech speed up the core vision pipeline behind its digital-shelf services. Data-processing time improved by more than 50%, DeepGPU accelerated vision inference by as much as 4.60x, GPU-cluster utilization improved 30%, and the platform kept 97%+ recognition accuracy while ensuring more than 95% of complex flows completed within 10 seconds.
Lenztech and Alibaba Cloud upgraded the AI product-recognition platform behind Trax China's digital-shelf services with a layered cloud AI architecture. At the data-preparation tier, CPFS provides high-throughput, low-latency file storage so many workers can write images in parallel, while OSS keeps storage economics under control and lets image data move between cost and performance tiers. For inference, Lenztech connected its self-developed high-performance operator library to the DeepGPU toolkit, accelerating multiple computer-vision models so recognition jobs can hit both high-accuracy and low-latency targets. Around the serving layer, the team combined its existing containerized architecture with Alibaba Cloud's cloud-native AI suite so GPU sharing, memory isolation, and flexible scheduling policies can keep inference tasks stable even as model versions and workloads change quickly. The result is a faster, more economical vision platform for shelf and retail analytics.
Retail images are written into CPFS and OSS, preprocessed for model input, and then sent to DeepGPU-accelerated computer-vision models on the cloud AI platform. The inference results are returned to retail analytics and shelf-monitoring services, while the cloud-native AI suite keeps GPU scheduling and model-serving resources stable as workloads and model versions evolve.
By combining its own operator library with DeepGPU, Lenztech accelerated multiple computer-vision models by 1.12x to 4.60x while serving daily recognition tasks. (original: 将上层模型算法推理能力与阿里云的 DeepGPU 工具包紧密结合... 成功实现了多个计算机视觉模型的推理加速，加速比区间达到 1.12 至 4.60 倍。)
Lenztech processes millions of images and massive SKU volumes every day, so the platform had to improve image handling speed and stay responsive in complex retail scenes. (original: 朗镜科技每日需处理数以百万计的图片及海量 SKU... 需要提升小文件处理能力，力求缩短下载延迟，实现即时响应。)
The cloud-native AI suite raised overall GPU-cluster utilization by 30% and reduced the operations complexity of the AI cluster. (original: 该方案深度融合了 GPU 共享调度、显存隔离、灵活的调度策略等技术，将现有 GPU 集群利用率整体提升了 30%，显著简化了 AI 集群的运维复杂度。)
Using CPFS for parallel writes reduced the share of small-image writes with more than 100 ms latency to under 5% and lifted overall data-processing speed by more than 50%. (original: 采用阿里云的高性能文件存储 CPFS... 小文件图片百毫秒以上写入延迟占比减少到 5% 以内，整体数据处理时间提升 50% 以上。)
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Alibaba Cloud Sample Oct 13, 2026 TOPSPORTS (滔搏) TOPSPORTS wanted to scale enterprise AI agents beyond isolated pilots, but every new agent risked recreating the same knowledge, data, tools, and skills in separate silos. Business-analysis, membership, and marketing… business analysis time reduced to 20 minutes AgentCore · AgentLoop · MCP tools Alibaba Cloud Sample Feb 23, 2025 Zhiyi Technology (知衣科技) Zhiyi Technology needed to keep a fashion-intelligence platform responsive while running large-scale image-recognition, vector retrieval, trend analysis, and recommendation workflows. Its self-built CDH environment was… the company completed a RMB 200 million Series B financing round in 2021 DataWorks · Elasticsearch · MaxCompute 
Google Cloud Sample Modified · Oct 9, 2026 Central Retail Food Replaced ~2,000 monthly manual Excel reports with a single source of truth on BigQuery through Looker dashboards Central Retail Food unifies its ecosystem on Google Cloud, moving from fragmented data to conversational… Addresses 50% to 60% of customer service queries with automated chatbots built on Conversational Agents BigQuery · Gemini · Gemini platform 