Upload folder using huggingface_hub

README.md

@@ -1,3 +1,111 @@
----
-license: apache-2.0
----
+# Chest CT Segmentation - Federated Learning Partitions
+
+Dataset partitions for **Federated SAM2-LoRA** medical image segmentation across multiple Data Owners.
+
+## Overview
+
+This dataset contains partitioned chest CT scans for federated learning with heterogeneous clients:
+
+| Data Owner | Type | Training Method | Contributes to FedAvg |
+|------------|------|-----------------|----------------------|
+| DO1 | Zero-shot | CLIP text prompts | No |
+| DO2 | Few-shot | Memory bank | No |
+| DO3 | LoRA | Gradient training | Yes |
+| DO4 | LoRA | Gradient training | Yes |
+
+## Statistics
+
+| Partition | Private (train/test) | Mock (train/test) | Patients |
+|-----------|---------------------|-------------------|----------|
+| do2_fewshot | 4 / 2 | 2 / 1 | 6 |
+| do3_lora | 16 / 4 | 8 / 2 | 20 |
+| do4_lora | 14 / 4 | 7 / 2 | 18 |
+
+- **Total images**: 44 private + 22 mock = 66 images
+- **Total patients**: 44 unique (1 image per patient, no overlap between DOs)
+- **Mock data**: Half of private data for testing/simulation
+
+## Structure
+
+```
+hf_partitions/
+├── do2_fewshot/
+│   ├── mock/
+│   │   ├── train/
+│   │   │   ├── images/
+│   │   │   ├── masks/
+│   │   │   └── train.csv
+│   │   └── test/
+│   │       ├── images/
+│   │       ├── masks/
+│   │       └── test.csv
+│   ├── private/
+│   │   ├── train/
+│   │   └── test/
+│   └── README.md
+├── do3_lora/
+│   ├── mock/
+│   ├── private/
+│   └── README.md
+├── do4_lora/
+│   ├── mock/
+│   ├── private/
+│   └── README.md
+└── README.md
+```
+
+## Usage
+
+### Download All Partitions
+```python
+from huggingface_hub import snapshot_download
+
+snapshot_download(
+    repo_id="khoaguin/chest-ct-segmentation",
+    repo_type="dataset",
+    local_dir="./dataset"
+)
+```
+
+### Download Specific DO Partition
+```python
+# Download only DO3 (LoRA) partition
+snapshot_download(
+    repo_id="khoaguin/chest-ct-segmentation",
+    repo_type="dataset",
+    allow_patterns="do3_lora/**",
+    local_dir="./dataset"
+)
+```
+
+### Download Mock Data Only (for testing)
+```python
+# Download mock data for all DOs
+snapshot_download(
+    repo_id="khoaguin/chest-ct-segmentation",
+    repo_type="dataset",
+    allow_patterns="*/mock/**",
+    local_dir="./dataset"
+)
+```
+
+## Data Format
+
+- **Images**: RGB JPEG chest CT slices
+- **Masks**: Binary segmentation masks (lung/tissue regions)
+- **CSV**: Mapping between ImageId and MaskId
+
+### CSV Format
+```csv
+ImageId,MaskId
+ID00131637202220424084844_30.jpg,ID00131637202220424084844_mask_30.jpg
+```
+
+## Related
+
+- [FL-SAM2-LoRA Project](https://github.com/OpenMined/fl-sam2-lora)
+- [Original Dataset](https://www.kaggle.com/datasets/polomarco/chest-ct-segmentation)
+
+## License
+
+Please refer to the original dataset license for usage terms.

do2_fewshot/README.md

@@ -0,0 +1,65 @@
+# DO2 Few-shot - Chest CT Segmentation
+
+Dataset partition for **Data Owner 2 (Few-shot)** in Federated SAM2-LoRA.
+
+## Client Configuration
+
+| Property | Value |
+|----------|-------|
+| Type | Few-shot |
+| Training Method | Memory bank (similarity-based) |
+| Gradient Training | No |
+| Contributes to FedAvg | No |
+
+## Statistics
+
+| Split | Mock | Private |
+|-------|------|---------|
+| Train | 2 (2 patients) | 4 (4 patients) |
+| Test | 1 (1 patient) | 2 (2 patients) |
+
+**Total: 6 unique patients (1 image per patient)**
+
+## Structure
+
+```
+do2_fewshot/
+├── mock/
+│   ├── train/
+│   │   ├── images/    (2 images)
+│   │   ├── masks/     (2 masks)
+│   │   └── train.csv
+│   └── test/
+│       ├── images/    (1 image)
+│       ├── masks/     (1 mask)
+│       └── test.csv
+├── private/
+│   ├── train/
+│   │   ├── images/    (4 images)
+│   │   ├── masks/     (4 masks)
+│   │   └── train.csv
+│   └── test/
+│       ├── images/    (2 images)
+│       ├── masks/     (2 masks)
+│       └── test.csv
+└── README.md
+```
+
+## Usage
+
+```python
+from huggingface_hub import snapshot_download
+
+# Download this partition
+snapshot_download(
+    repo_id="khoaguin/chest-ct-segmentation",
+    repo_type="dataset",
+    allow_patterns="do2_fewshot/**",
+    local_dir="./dataset"
+)
+```
+
+## Related
+
+- [DO3 LoRA](../do3_lora/) - Full LoRA training (16 train, 4 test)
+- [DO4 LoRA](../do4_lora/) - Full LoRA training (14 train, 4 test)

do2_fewshot/mock/test/test.csv

@@ -0,0 +1,2 @@
+ImageId,MaskId
+ID00140637202231728595149_30.jpg,ID00140637202231728595149_mask_30.jpg

do2_fewshot/mock/train/train.csv

@@ -0,0 +1,3 @@
+ImageId,MaskId
+ID00131637202220424084844_30.jpg,ID00131637202220424084844_mask_30.jpg
+ID00249637202266730854017_190.jpg,ID00249637202266730854017_mask_190.jpg

do2_fewshot/private/test/test.csv

@@ -0,0 +1,3 @@
+ImageId,MaskId
+ID00140637202231728595149_30.jpg,ID00140637202231728595149_mask_30.jpg
+ID00273637202271319294586_39.jpg,ID00273637202271319294586_mask_39.jpg

do2_fewshot/private/train/train.csv

@@ -0,0 +1,5 @@
+ImageId,MaskId
+ID00131637202220424084844_30.jpg,ID00131637202220424084844_mask_30.jpg
+ID00249637202266730854017_190.jpg,ID00249637202266730854017_mask_190.jpg
+ID00014637202177757139317_21.jpg,ID00014637202177757139317_mask_21.jpg
+ID00170637202238079193844_153.jpg,ID00170637202238079193844_mask_153.jpg

do3_lora/README.md

@@ -0,0 +1,67 @@
+# DO3 LoRA - Chest CT Segmentation
+
+Dataset partition for **Data Owner 3 (LoRA)** in Federated SAM2-LoRA.
+
+## Client Configuration
+
+| Property | Value |
+|----------|-------|
+| Type | LoRA |
+| Training Method | Gradient-based fine-tuning |
+| LoRA Rank | 16 |
+| Learning Rate | 1e-4 |
+| Augmentation | Enabled |
+| Contributes to FedAvg | Yes |
+
+## Statistics
+
+| Split | Mock | Private |
+|-------|------|---------|
+| Train | 8 (8 patients) | 16 (16 patients) |
+| Test | 2 (2 patients) | 4 (4 patients) |
+
+**Total: 20 unique patients (1 image per patient)**
+
+## Structure
+
+```
+do3_lora/
+├── mock/
+│   ├── train/
+│   │   ├── images/    (8 images)
+│   │   ├── masks/     (8 masks)
+│   │   └── train.csv
+│   └── test/
+│       ├── images/    (2 images)
+│       ├── masks/     (2 masks)
+│       └── test.csv
+├── private/
+│   ├── train/
+│   │   ├── images/    (16 images)
+│   │   ├── masks/     (16 masks)
+│   │   └── train.csv
+│   └── test/
+│       ├── images/    (4 images)
+│       ├── masks/     (4 masks)
+│       └── test.csv
+└── README.md
+```
+
+## Usage
+
+```python
+from huggingface_hub import snapshot_download
+
+# Download this partition
+snapshot_download(
+    repo_id="khoaguin/chest-ct-segmentation",
+    repo_type="dataset",
+    allow_patterns="do3_lora/**",
+    local_dir="./dataset"
+)
+```
+
+## Related
+
+- [DO2 Few-shot](../do2_fewshot/) - Few-shot learning (4 train, 2 test)
+- [DO4 LoRA](../do4_lora/) - Full LoRA training (14 train, 4 test)

do3_lora/mock/test/test.csv

@@ -0,0 +1,3 @@
+ImageId,MaskId
+ID00111637202210956877205_179.jpg,ID00111637202210956877205_mask_179.jpg
+ID00255637202267923028520_25.jpg,ID00255637202267923028520_mask_25.jpg

do3_lora/mock/train/train.csv

@@ -0,0 +1,9 @@
+ImageId,MaskId
+ID00102637202206574119190_84.jpg,ID00102637202206574119190_mask_84.jpg
+ID00047637202184938901501_20.jpg,ID00047637202184938901501_mask_20.jpg
+ID00035637202182204917484_562.jpg,ID00035637202182204917484_mask_562.jpg
+ID00167637202237397919352_195.jpg,ID00167637202237397919352_mask_195.jpg
+ID00285637202278913507108_1.jpg,ID00285637202278913507108_mask_1.jpg
+ID00207637202252526380974_40.jpg,ID00207637202252526380974_mask_40.jpg
+ID00267637202270790561585_8.jpg,ID00267637202270790561585_mask_8.jpg
+ID00367637202296290303449_218.jpg,ID00367637202296290303449_mask_218.jpg