Create README_Brain-Tumor_MedThinkSeg.md

Brain-Tumor/README_Brain-Tumor_MedThinkSeg.md

@@ -0,0 +1,125 @@
+# Brain-Tumor (Kaggle “Brain Tumor Image DataSet : Semantic Segmentation”) — packaged in `saiteja33/MedThink-Seg/Brain-Tumor`
+
+This README documents:
+1) **The original Kaggle Brain Tumor Image DataSet : Semantic Segmentation** (what it contains, size, and original structure), and  
+2) **How it is organized inside this Hugging Face repo** (`saiteja33/MedThink-Seg`, under the `Brain-Tumor/` directory).
+
+---
+
+## 1) Original dataset (source)
+
+**Original Kaggle dataset:**  
+https://www.kaggle.com/datasets/pkdarabi/brain-tumor-image-dataset-semantic-segmentation
+
+**Creator (as cited in dataset cards mirroring the Kaggle release):** Peyman Darabi (Aug 2023)
+
+### What kind of images are included?
+- **Modality:** Brain **MRI** images for tumor region analysis.
+- **Image format & size (commonly reported for this Kaggle release):**
+  - **JPEG** images
+  - **640 × 640** resolution
+- **Classes:** 2 classes:
+  - **Tumor**
+  - **Non-tumor / Normal**
+
+### Dataset size and counts
+A widely-used mirror of this Kaggle dataset reports:
+- **Total images:** **2,146**
+- **Splits:**
+  - **Train:** 1,502
+  - **Validation:** 429
+  - **Test:** 215
+
+> Note: The original Kaggle page is sometimes hard to scrape automatically; the counts above match publicly available mirrors of the Kaggle dataset.
+
+### Labels and annotations (original)
+The original release is commonly distributed as **COCO-style annotations**, including:
+- **Bounding boxes** (`bbox`)
+- **Polygon segmentation coordinates** (`segmentation`)
+
+Some downstream analyses report that many “segmentations” are effectively **rectangular polygons** (i.e., bounding boxes), so this dataset may behave more like a detection dataset unless masks are rasterized carefully.
+
+### Original naming convention (examples)
+Mirrors of the dataset show filenames like:
+- `2256_jpg.rf.<hash>.jpg`
+
+(Your copy may preserve these names.)
+
+---
+
+## 2) Organization inside this Hugging Face repo (`saiteja33/MedThink-Seg`)
+
+**HF path:** `Brain-Tumor/`  
+**Repo UI size (shown in HF browser):** ~**466 MB**
+
+This packaging contains **only images and masks** (no separate `normal/` folder).
+
+### Directory structure (this repo)
+```
+Brain-Tumor/
+├─ images/        # MRI images
+└─ masks/         # segmentation masks (paired with images/)
+```
+
+### Pairing rule
+- Each file in `Brain-Tumor/images/` has a corresponding mask in `Brain-Tumor/masks/`.
+- **Filename matching:** mask filename matches the image filename (same base name).  
+  Example (conceptual):
+  - `images/<name>.jpg`  ↔  `masks/<name>.png`  (or same extension if you kept it consistent)
+
+### Mask format (practical note)
+- If masks were rasterized from COCO polygons, they are typically **binary** masks:
+  - background = 0
+  - tumor = 1 (or 255)
+- Verify a few masks to confirm pixel values before training.
+
+---
+
+## How to download / use (recommended)
+
+### A) Download only `Brain-Tumor/` from this HF repo
+```python
+from huggingface_hub import snapshot_download
+
+local_dir = snapshot_download(
+    repo_id="saiteja33/MedThink-Seg",
+    repo_type="dataset",
+    allow_patterns=["Brain-Tumor/**"],  # only pull Brain-Tumor subdir
+)
+print(local_dir)
+```
+
+### B) Create (image, mask) pairs
+```python
+from pathlib import Path
+
+root = Path(local_dir) / "Brain-Tumor"
+img_dir = root / "images"
+msk_dir = root / "masks"
+
+images = sorted(list(img_dir.glob("*")))
+pairs = [(p, msk_dir / p.name) for p in images]
+
+missing = [p.name for p, m in pairs if not m.exists()]
+print("num_images:", len(images))
+print("missing_masks:", len(missing))
+print("example_pair:", pairs[0] if pairs else None)
+```
+
+### C) Load with 🤗 Datasets (images only)
+```python
+from datasets import load_dataset
+
+ds_imgs = load_dataset("imagefolder", data_dir=str(img_dir))
+print(ds_imgs)
+```
+
+(For segmentation training, most pipelines keep pairs via the folder structure as in **B**.)
+
+---
+
+## Citation
+If you use this dataset, cite the Kaggle record:
+
+- Peyman Darabi. *Brain Tumor Image DataSet : Semantic Segmentation.* Kaggle Dataset.  
+  https://www.kaggle.com/datasets/pkdarabi/brain-tumor-image-dataset-semantic-segmentation