FEMNIST

Dataset Overview

1. FEMNIST

• Type: Image dataset
• Details: Contains 62 classes (10 digits, 26 lowercase letters, 26 uppercase letters), image size 28×28 pixels, optionally adjustable to 128×128 pixels, with a total of 3,500 users.
• Task: Image classification

2. Sentiment140

• Type: Text dataset containing tweets
• Details: 660,120 users
• Task: Sentiment analysis
• Format: Includes training and test sets; data stored in JSON format, each user’s data includes tweet content and sentiment label.

3. Shakespeare

• Type: Text dataset containing Shakespeare dialogues
• Details: 1,129 users (reduced to 660 users)
• Task: Next‑character prediction
• Format: Text format containing dialogue content.

4. Celeba

• Type: Image dataset
• Details: 9,343 users (excluding celebrities with fewer than 5 images)
• Task: Image classification (smiling vs. non‑smiling)

5. Synthetic Dataset

• Type: Synthetic dataset
• Details: Users can customize the number of devices, number of classes, dimensions, etc.
• Task: Classification

6. Reddit

• Type: Text dataset containing Reddit comments
• Details: 1,660,820 users, total of 56,587,343 comments
• Task: Next‑word prediction

7. CIFAR 10 / CIFAR 100

• Type: Image classification dataset
• Details: 60,000 color images of size 32×32 pixels, distributed across 10 and 100 classes respectively, with 50,000/10,000 training/testing split.
• Task: Image classification

8. FedVision - Street Dataset

• Type: Real‑world object detection dataset
• Details: Contains 5,20 devices, 956 samples, 7 classes
• Task: Object detection
• Format: Includes image data and training labels, stored in JSON format.

9. EMNIST

• Type: Extended MNIST dataset containing English letters and digits
• Details: Divided into 6 subsets: By_Class, By_Merge, Balanced, Digits, Letters, and MNIST
• Task: Classification

10. MovieLens

• Type: Structured dataset
• Details: Contains user ratings for videos and video attributes; ratings are on a 5‑point scale
• Task: Recommendation system
• Format: Includes ratings.dat, users.dat, and movies.dat

11. Credit

• Type: Structured dataset
• Details: Contains user attributes such as gender, education level, etc. Credit 1 includes 150,000 samples with 10 attributes; Credit 2 includes 30,000 samples with 25 attributes
• Task: Classification (predict whether a user will default on repayment)

12. ModelNet

• Type: Image classification dataset
• Details: Contains 2,311 3D models from 40 categories captured from various viewpoints
• Task: Image classification
• Processing: Requires conversion of CAD models to images using open‑source software Blender

13. PersonaChat

• Type: Dialogue dataset
• Details: Naturally non‑i.i.d. partitioned, based on assigned personas, divided into 17,568 clients

14. KWS

• Type: Speech command dataset
• Task: Limited‑vocabulary speech recognition

15. Flickr

• Type: Personalized image aesthetic dataset
• Task: Personalized image classification