#Japan #AI #Tokyo #Uridoki #ImageClassifier

Enhancing User Experience with AI at Uridoki

Uridoki, a consumer-to-business (C2B) trading platform based in Tokyo, has recently announced the introduction of its new AI model, the Uridoki Item Image Classifier. This innovative technology is set to streamline the process of soliciting appraisals for pre-owned items, making it simpler and more efficient for users. With this development, Uridoki aims to enhance the overall service experience by utilizing AI technology to assist in automatic category selection, ultimately reducing the burden on users during the appraisal request process.

How the AI Model Works

The Uridoki Item Image Classifier operates by allowing users to upload images of their items when requesting an appraisal. The AI then analyzes these images to determine the likelihood of the items belonging to various categories, presenting this data as a probability distribution. If the probability for a specific category is significantly higher than others, the system will classify the item accordingly. However, in cases where the distributions are less distinct, users receive a selection of candidate categories to choose from, thus guiding them to make appropriate selections.

The AI has been specifically optimized to accurately distinguish images of items commonly seen in the re-use industry, such as the bottoms and insides of brand bags or high-end watches taken from specific angles.

Extensive Training Data

This model has been developed using a substantial dataset comprising over 1.2 million real images of pre-owned items that have been collected by Uridoki. Currently, it can categorize items in 42 distinct categories, achieving an impressive overall classification accuracy of 87%. When focusing solely on frontal images and full views submitted by users during appraisal requests, the classification accuracy exceeds 95%, thus confirming the model's effectiveness in real-world applications.

Addressing User Challenges

Historically, users have been required to select the category manually when submitting an appraisal request, a process that often led to confusion and frustration. With the introduction of the Uridoki Item Image Classifier, users can bypass this meticulous step, which contributes significantly to a more user-friendly experience on the platform.

The necessity for manual selection prompted Uridoki to innovate, leading to the creation of a unique image recognition model that automatically categorizes items based on images uploaded by users. As a result, Uridoki enhances convenience and decreases input workload, allowing users to focus on other important aspects of their selling process.

Unique Features of the AI Model

The AI model distinguishes itself from standard image recognition systems by focusing solely on the re-use sector. In real-life transactions, users frequently post images of items showcasing only parts of products or those that depict usage wear and tear. This AI model is specifically fine-tuned to recognize such specific angles and conditions, driving superior categorization performance even in challenging scenarios.

Processing Flow of the AI Model

The operational flow of the AI model consists of the following steps:
1. Image Upload: Users upload images of the items they wish to have appraised via smartphones or PCs.
2. Automated Analysis: The AI instantaneously processes the uploaded image data to classify the item.
3. Category Assignment: Rather than confining images to a single category, the model generates a probability distribution for each category, identifying the most probable category for the item.
4. Handling Complex Cases: For images where probabilities are spread across multiple categories, the system provides a fallback mechanism that allows users to choose directly from a list of suggested categories.

Future Developments

Uridoki is committed to leveraging AI technologies to continually refine and enhance its services. This is not the first AI initiative from Uridoki. Previous projects include the automatic detection of abnormal appraisal prices, the deployment of an appraisal image analysis system, and the launch of a review moderation system.

The Uridoki Item Image Classifier marks the fourth installment in this series of AI endeavors aimed at improving user experiences within the platform. As Uridoki accumulates more data through ongoing operations, the platform is set to continue learning and improving its classification accuracy to adapt to evolving trends and product appearances in the re-use marketplace.

Continuous Learning and Improvement

Establishing a feedback mechanism that incorporates the latest image data and assessment outcomes into the AI model is a critical part of Uridoki's strategy. This adaptable system ensures responsiveness to new products entering the market, shifts in user behavior, and unique challenges posed by the re-use sector.

By capitalizing on daily operational insights, Uridoki aims to advance the practical application of AI technologies and enhance features in the re-use space. The company remains dedicated to delivering a high-quality platform that allows users to sell their goods more easily and securely.

For more information on Uridoki and to experience this innovative platform, visit Uridoki's Website.