DiveDeepAI

Card Vision

The experts at DivedeepAI developed Card Vision, a smart computer vision operated application that is used to determine the condition, authenticity of a Pokemon Card.

About Client

A stealth startup focused on leveraging Computer Vision to solve real world problems.

The Challenge

The main Challenge was data collection and information extraction. There was no subgrades (edges, surface etc.) information available with cards on the ebay website so we had to manually write up all this information for Models’ training. If we did not do this, we would not be able to predict card conditions. Moreover, we had to crop and preprocess all images because these cards were laminated and we had to preprocess these for text and other features extraction such as damage detection, border thickness check etc.

The Solution

We used deep learning model ResNet for training and prediction of these subgrade values to determine card condition. After completing data training, We saved the trained ResNet model for future use to predict a card’s grade, corner, edges, centering and surface condition. To determine the authenticity of a pokemon card we implemented a check of 8 factors using different python modules such as openCV, Pytesseract, and numpy etc.

The Impact

The application checks for these features in each card and tells which tests pass and which one fails. If all 8 factors pass the checks then the card is real. Even if one check fails, we call it an error card.

HIGHLIGHTS

Project in Action

Preprocessing Data

Training ResNet

Grade Prediction

Condition Assessment

Client Testimonial

Amazing Dev! Do not hesitate to hire! Finished the job promptly and made sure all my needs were met!