All the training images are uploaded in the train folder with prefix Image and the respective class labels have been mapped in train_label.csv. In train_label.csv, we have normalized the target classes to 0-6(7 classes in total).
Context:
Attributes are important characteristics of the fashion products that define their uniqueness. For instance, long sleeve length and round collar are two attributes (among many) that characterize shirts. Likewise there are numerous attributes defining a certain product and the uniqueness associated with it.
Extracting relevant attributes from fashion images is crucial to driving user demand and preferences and solving fashion-related problems.
Challenges The data is tagged internally and has class imbalance. The accuracy of the tags aren't ideal and have few errors (noise). The problem requires participants to come up with efficient training strategies to overcome these challenges.
to run this code You can visit my notebook in kaggle:
https://www.kaggle.com/code/abhishek725kumar/notebookc239a6fa51