Update README.md

README.md

@@ -13,12 +13,53 @@ Binary image dataset indicating whether an image **contains a Baby Yoda LEGO** f
 - `has_baby_yoda` (label=0)
 - `no_baby_yoda` (label=1)
 
+## Composition & Collection
+- **Creators:** Student-captured photos by the dataset author for an academic assignment.
+- **Subjects:** Desktop/object scenes that may contain a Baby Yoda LEGO figure.
+- **Original count:** {orig_n} images (≥30 required).
+- **Resolution / format:** Center-cropped square, resized to **224×224**, saved as JPEG.
+- **Privacy:** No faces or personally identifiable information (PII) included.
+
+## Labels
+Binary target from folder names:
+- `has_baby_yoda` (label = 0)
+- `no_baby_yoda` (label = 1)
+
+
+
+## Preprocessing
+1. Decode `.heic/.jpg/.jpeg/.png` (HEIC via `pillow-heif`).
+2. Center-crop to square using `min(width, height)`.
+3. Resize to **224×224** with bilinear interpolation.
+4. Save as JPEG (quality≈92).
+
+## Augmentation (label-preserving)
+Performed offline to create the `augmented` split (no generative models):
+- `RandomResizedCrop(size=224, scale=(0.7, 1.0), ratio=(0.75, 1.33))`
+- `RandomHorizontalFlip(p=0.5)`
+- `RandomVerticalFlip(p=0.1)`
+- `RandomRotation(±15°, bilinear)`
+- `ColorJitter(brightness=0.2, contrast=0.2, saturation=0.15, hue=0.05)`
+- `RandomErasing(p=0.2, scale=(0.02, 0.08), ratio=(0.3, 3.3))`
+
+These transformations do not change whether Baby Yoda is present.
+
 ## Splits
-- `original`: 30 resized images (224×224)
-- `augmented`: 300 label-preserving synthetic images
+- `original`: {orig_n} resized images (224×224)
+- `augmented`: {aug_n} synthetic, label-preserving variants (≥300 total)
+
+## Intended Use / Limits
+- **Use:** Classroom demos, practice with CV pipelines, small experiments.
+- **Not for:** Benchmarks or production systems (limited scope/diversity).
+- **Caveats:** Background/lighting vary; class balance may be modest.
+
+## Ethical Notes
+- No people or PII.
+- Images are student-created and used for coursework.
 
-## Augmentations
-RandomResizedCrop, flips, rotation, color jitter, sharpness, autocontrast, and random erasing.
+## Licensing
+- **CC-BY-4.0** for the dataset and documentation. Please credit the dataset if reused.
 
-## Notes
-Images are center-cropped to square and resized to 224×224. Labels come from folder names.
+## AI Usage Disclosure
+- No generative AI to produce new originals.
+- Augmented split created with standard `torchvision` transforms only.