Telegram Stickers Image Classification Dataset

This dataset consists of a collection of Telegram stickers that have been converted into images for the purpose of image classification.

Dataset Details

• Image Size: 512x512 pixels
• Number of Classes: 1276
• Total Number of Images: 672,911

The dataset was created by extracting stickers from 23,681 sets of stickers in Telegram. Animated and video stickers were removed, and sets that had only one emoji assigned to all stickers were ignored. Stickers that did not fit the 512x512 size were padded with empty pixels. Furthermore, all stickers were converted to the .png format to ensure consistency.

The class names for the stickers were assigned based on the Unicode emoji given to them by the author. For example, the Unicode U+1F917 represents the 🤗 emoji. Each sticker in the dataset is labeled with the corresponding Unicode code as its class.

The name of each image in the dataset corresponds to the file ID of the sticker in Telegram. This unique identifier can be used to reference the original sticker in the Telegram platform.

Dataset Split

• Training Set:

  • Number of Images: 605,043

• Validation Set:

  • Number of Images: 33,035

• Test Set:

  • Number of Images: 34,833

Additional Information

The training set train.zip has been divided into multiple parts, each of which is approximately 20 GB in size. To extract the dataset, you will need a program that supports extracting split archives, such as 7z.

In the dataset_resized folder, you will find the resized version of the dataset. The images in this folder have been resized to 128x128 pixels.

Please note that the original dataset provided is in the format of 512x512-pixel images, while the dataset_resized folder contains the resized images of 128x128 pixels.