## Files

*   **train.csv** - the training set
*   **/train** - directory of images for the training set
*   **test.csv** - the test set
*   **/test** - directory of images for the test set
*   **sample_submission.csv** - a sample submission file in the correct format

## Columns

*   `id` - the file name without the extension
*   `class` - Class of image for example "20 Hellers,Czech Koruna,czech_republic"
*   etc.


- **Train**: ~10,400 images (85% of the dataset).
- **Test**: ~1,288 images (15% of the dataset).

Each image is labeled with one of **315 classes**, combining **country + currency + denomination** into a single identifier. The dataset was assembled through:
- **Web scraping** across various coin-collector resources
- The coin image dataset [WanderDust’s Coin Images](https://www.kaggle.com/datasets/wanderdust/coin-images/data)

### Challenges & Caveats

- **Class Imbalances**: Some coins have many samples, while others are relatively underrepresented.
- **Mint Variations**: Coins of the same denomination might differ subtly due to different mints or yearly revisions.
- **Type Variations**: Coins of the same denomination might have different types.
- **Varied Conditions**: Tarnish, rust, and scratches can alter a coin’s appearance. Lighting and angle variations further increase complexity.
- **Subtle Differences**: Denomination of euro coins can only be distinguished by edge patterns.