SKU110K-DenseDet : The machine learning model can detect products within a store.

elegirl_admin

ailia.AI

ailia.AI Editorial Team

Here’s the translation:

Introducing “SKU110K-DenseDet,” a machine learning model available for use with the ailia SDK. By utilizing machine learning models published on ailia MODELS and the ailia SDK, an edge-oriented inference framework, you can easily implement AI functionality into your applications.

Overview of SKU110K-DenseDet

SKU110K-DenseDet is a machine learning model capable of detecting products in store environments. It can detect bounding boxes of products from supermarket shelves, without categorization, solely based on whether the product is present or not.

Source: GitHub Repository

SKU110K Dataset

SKU110K is a dataset for product detection released in April 2019. It includes images captured from mobile phones in over 1000 supermarkets worldwide (United States, Europe, East Asia). Bounding boxes are manually annotated, with 8233 images containing 90968 bounding boxes for training and 2941 images containing 432312 bounding boxes for validation.

Source: Paper

SKU110K-DenseDet Architecture

SKU110K-DenseDet is trained using MMDetection, utilizing Cascade R-CNN and achieving 58.0% mAP. It uses ResNeXt-101 as the backbone.

Due to SKU110K containing many small objects, conventional object detection architectures and recognition resolutions do not yield satisfactory accuracy. Therefore, the dataset is rescaled to 2560×2560 for training.

Additionally, considering environments with limited GPU memory where detection by cropping is necessary, random crop is used during training.

SKU110K contains an average of 150 bounding boxes per image, significantly more than MS COCO. Default hyperparameters do not yield satisfactory accuracy, so adjustments are made to the max positive sample number for RPN and R-CNN.

How to Use SKU110K-DenseDet

To use SKU110K-DenseDet with the ailia SDK, use the following command. Due to the large backbone, in environments with limited GPU memory, specify -e 0 to execute in CPU mode.

$ python3 sku110k-densedet.py -i input.jpg -e 0

ax Inc. is a company specializing in practical AI solutions, developing the ailia SDK for cross-platform high-speed inference using GPUs. From consulting to model creation, SDK provision, AI-based application/system development, and support, ax Inc. provides a comprehensive solution for AI-related needs. Please feel free to contact us.

Source: GitHub Repository