Back to datasets
Dataset assetOpen Source CommunityImage ProcessingHandwritten Digit Recognition
HwD-1000
The dataset contains 1,000 handwritten digit images, each 28 × 28 pixels with a white background and black strokes, covering the digits 0–9.
Source
github
Created
Sep 21, 2024
Updated
Oct 8, 2024
Signals
156 views
Availability
Linked source ready
Overview
Dataset description and usage context
Handwritten Digit Recognition
Dataset Overview
- Name: HwD‑1000 dataset
- Link: https://github.com/niklashenning/hwd-1000-dataset
- Description: Contains 1,000 images of handwritten digits (0‑9). Images are 28 × 28 pixels, white background, black pen strokes, with varying styles and thicknesses.
Usage
- Training Models: Train neural networks using PyTorch and the Tensor library.
- Data Processing: Load and process the dataset with pandas, load and transform images with Pillow, visualize training results with Matplotlib.
Training Results
- Training Settings:
- Epochs: 50
- Learning Rate: 0.001
- Optimizer: AdamW
- Loss Function: CrossEntropyLoss
- Split: 80 % train, 20 % validation
- Validation Accuracy: 99.50 %
- Test Accuracy: 99.50 % (199/200)
Training Loss
| Epoch | Loss |
|---|---|
| 1 | 1.506903 |
| 10 | 0.047416 |
| 20 | 0.011299 |
| 30 | 0.006869 |
| 40 | 0.002777 |
| 50 | 0.001392 |
License
- License Type: MIT license
Need downstream help?
Pair the dataset with AI analysis and content workflows.
Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.