# 180DC - ML SIG Recruitment — Data

来源：https://www.kaggle.com/competitions/180-dc-ml-sig-recruitment/data

采集时间：2026-10-19T10:36:47Z

The dataset, `REC_DATASET`, is designed for a **Supervised Image Denoising + Classification** task.
It contains flower images from 5 categories:

1. **Daisy**
2. **Dandelion**
3. **Roses**
4. **Sunflowers**
5. **Tulips**

### **Structure**

#### **Train Set (`train/`)**

* Organized into subfolders by class (`1`, `2`, `3`, `4`, `5`).
* Each subfolder contains **paired images**:

  * **Clean images** (original)
  * **Noisy images** (with partial Gaussian noise)
* Filenames follow the pattern:

  * `Image_Train_<id>.png` for noisy images
  * `Image_Clean_<id>.png` for clean images

#### **Test Set (`test/`)**

* Contains only **noisy images**, randomly shuffled.
* Filenames follow the pattern:

  * `Image_Test_<id>.png`
* No clean images are provided in the test set.

#### **Labels File (`test_labels.csv`)**

* Contains the ground truth class for each test image (hidden from participants during the competition).
* Format:

  ```
  Images,Classes
  Image_Test_1.png,3
  Image_Test_2.png,5
  ...
  ```

---

This dataset setup ensures that:

* You can **train** a supervised denoising model using clean-noisy pairs from the training set.
* You must **generalize** to noisy images in the test set without direct access to their clean counterparts.

---