资源算法Identify-Characters-From-Product-Images

Identify-Characters-From-Product-Images

2020-01-06 | |  57 |   0 |   0

This project was done as a part of the Identify Characers From Product Images competition. The competition was hosted in CrowdAnalytix.

Data (Provided by CrowdAnalytix)

The data consisted of product images like t-shirts, bags, keychains, mobile covers, etc. with characters graphics.

Training Set: 6694 images across 42 categories Test Set: 3727 images

The data had label inconsistency of 2-3%. Had to manually resolve the label inconsitency.

Additional Data

As the data provided by CrowdAnalytix was not equally distributed across the 42. Some categories had fewer data comparatively to the other categories. To resolve this data insufficiency among the categories we downloaded the additional data with the help of gi2ds  and this tutorial created by Andrian Rosebrook .

You can download the data from here (Comprises the data provided by CrowdAnalyticsX and the above mentioned additional data). The filenames of images from Crowdanalytix starts with Cax_train and the other images filenames start with number.

Categories

The following were the 42 categories for classification.

  • Angry Birds

  • Baloo

  • Bart simpson

  • Ben

  • Bulbasaur

  • Charizard

  • Charlie brown

  • Charmender

  • Chicken_little

  • Cinderella

  • Darth_vader

  • Disney_princes

  • Donald_duck

  • Godzilla

  • Goku

  • Goofy

  • Han-solo

  • Harry_potter

  • Hellokitty

  • Itachi

  • John_Cena

  • Jojosiwa

  • Kakashi

  • Marilyn_monroe

  • Mickey_mouse

  • Minions

  • Naruto

  • Pikachu

  • Pokemon

  • Popeye

  • Power_rangers

  • R2-D2

  • Roman_reigns

  • Scoopy Doo

  • SpongeBob SquarePants

  • Squirtle

  • Teenage_mutant_ninja_turtles

  • Tom and Jerry

  • Toy_story_characters

  • Vampirina

  • Vegeta

  • Winnie the poo

Categories

Confusion Matrix

Confusion Matrix

Pair of confused categories with minimum value of 2

[('Cinderella', 'disney_princes', 6), ('pokemon', 'pikachu', 6), ('Baloo', 'Godzilla', 5), ('Charlie brown', 'goofy', 5), ('Charlie brown', 'Bart simpson', 4), ('popeye', 'power_rangers', 4), ('Roman Reigns', 'han-solo', 3), ('Scoopy Doo', 'Bart simpson', 3), ('disney_princes', 'Baloo', 3), ('disney_princes', 'Cinderella', 3), ('donald_duck', 'goofy', 3), ('goofy', 'mickey_mouse', 3), ('popeye', 'goofy', 3), ('vampirina', 'jojosiwa', 3), ('Bart simpson', 'Charlie brown', 2), ('Charlie brown', 'teenage_mutant', 2), ('Chicken_little', 'power_rangers', 2), ('Godzilla', 'Tom and Jerry', 2), ('Godzilla', 'goofy', 2), ('Goku_1', 'goofy', 2), ('Goku_1', 'pokemon', 2), ('John Cena', 'Roman Reigns', 2), ('Tom and Jerry', 'goofy', 2), ('charizard', 'Goku_1', 2), ('disney_princes', 'Bart simpson', 2), ('harry_potter', 'han-solo', 2), ('itachi', 'Goku_1', 2), ('kakashi', 'power_rangers', 2), ('naruto', 'Goku_1', 2), ('naruto', 'pokemon', 2), ('pikachu', 'pokemon', 2), ('pokemon', 'angrybirds', 2), ('popeye', 'mickey_mouse', 2), ('squirtle', 'bulbasaur', 2), ('squirtle', 'power_rangers', 2), ('squirtle', 'teenage_mutant', 2), ('vegeta', 'goofy', 2)]

Sample Images from the confused categories

Misclassified Images

Accuracy

The validation accuracy obtained was 90.7%

Leaderboard Ranking

The public leaderboard ranking: 12th with test accuracy of 89%

PublicLeaderboard

The private leaderboard ranking: 11th with test accuracy of 90%

Private Leaderboard


上一篇:Sports-Type-Classifier

下一篇:IDADataChallenge

用户评价
全部评价

热门资源

  • Keras-ResNeXt

    Keras ResNeXt Implementation of ResNeXt models...

  • seetafaceJNI

    项目介绍 基于中科院seetaface2进行封装的JAVA...

  • spark-corenlp

    This package wraps Stanford CoreNLP annotators ...

  • capsnet-with-caps...

    CapsNet with capsule-wise convolution Project ...

  • inferno-boilerplate

    This is a very basic boilerplate example for pe...