Upload folder using huggingface_hub

README.md

@@ -1,77 +1,38 @@
----
-tags:
-- ml-intern
----
-# KYC Document Corner Detector
+# kyc-document-corner-detector
 
-Lightweight document segmentation model trained on KYC documents (Aadhaar, PAN, passports, visas) from the `Jwalit/moire-docs` dataset.
+KYC Document Segmentation Model | MobileNetV3-Small | CPU Trained
 
-## Model Details
+## Details
+- **Task**: document_segmentation
+- **Backbone**: mobilenet_v3_small
+- **Input size**: 224px
+- **Epochs**: 8
+- **Best metric**: 0.8262 IoU
+- **Dataset**: Jwalit/moire-docs
+- **Total images**: 2623
 
-| Property | Value |
-|----------|-------|
-| Architecture | MobileNetV3-Small encoder + upsampling decoder |
-| Task | Binary segmentation (document vs background) |
-| Training | CPU only, 8 epochs |
-| Images | 461 (391 train, 70 val) |
-| **Best Val IoU** | **74.79%** |
-| Model size | ~10 MB |
-| Labels | Self-supervised via OpenCV contour detection |
+## Training
 
-## How It Works
-
-1. **Input**: Raw KYC document image (any size)
-2. **Segmentation**: Model predicts binary mask of document region
-3. **Corner Detection**: OpenCV contour extraction finds 4 corners from mask
-4. **Perspective Transform**: Crops to document boundaries
-
-## Self-Supervised Label Generation
-
-Labels are generated automatically using classical computer vision:
-- Grayscale → Gaussian blur → Adaptive thresholding
-- Morphological closing connects text regions
-- Largest contour extraction → 4-corner approximation
-
-No manual annotation required.
-
-## Files
-
-| File | Description |
-|------|-------------|
-| `pytorch_model.bin` | Trained model weights |
-| `config.json` | Model configuration |
-| `inference_pipeline.py` | Complete inference script (crop + rotate) |
-| `train_rotation_classifier.py` | Script to train rotation classifier |
+This model was trained on CPU using self-supervised labels:
+- **Segmentation**: OpenCV-generated document masks
+- **Rotation**: Synthetically rotated with known angles
 
 ## Usage
 
 ```python
 import torch
-from inference_pipeline import SegModel, predict_corners
+from model import YourModelClass  # See training script
 
-model = SegModel()
+model = YourModelClass()
 model.load_state_dict(torch.load("pytorch_model.bin", map_location="cpu"))
-corners = predict_corners(model, "your_document.jpg")
+model.eval()
 ```
 
-## Related Model
-
-- **Rotation Classifier**: https://huggingface.co/Jwalit/kyc-document-rotation-classifier
-  - 4-class classifier: 0°, 90°, 180°, 270°
-  - Run `train_rotation_classifier.py` to train on your CPU
-
-## Pipeline
-
-```
-Raw Image → SegModel → Mask → Contours → 4 Corners → Crop
-                                    ↓
-                              RotModel → Classify Rotation → Correct Orientation
-```
+## Dataset
 
-<!-- ml-intern-provenance -->
-## Generated by ML Intern
+- Source: `Jwalit/moire-docs`
+- Contains KYC documents with clean and moire (scan artifacts) variants
 
-This model repository was generated by [ML Intern](https://github.com/huggingface/ml-intern), an agent for machine learning research and development on the Hugging Face Hub.
+## License
 
-- Try ML Intern: https://smolagents-ml-intern.hf.space
-- Source code: https://github.com/huggingface/ml-intern
+Same as dataset license.

config.json

@@ -1,6 +1,9 @@
 {
-  "task": "segmentation",
+  "task": "document_segmentation",
   "backbone": "mobilenet_v3_small",
   "img_size": 224,
-  "epochs": 8
+  "epochs": 8,
+  "device": "cpu",
+  "best_metric": "0.8262 IoU",
+  "num_images": 2623
 }

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c469834f4850e6e69d9e15c783ef85f8d9251f54413b2e78a2e142a372df2b7
-size 10718163
+oid sha256:5965be686006a0b869f02cc3dc1f1222fb28d98c89bb8f1dd67e21aa1b1db7d9
+size 10735543

training_history.json

@@ -0,0 +1,58 @@
+[
+  {
+    "epoch": 1,
+    "train_loss": 0.4692644519617908,
+    "val_loss": 0.34619138449430464,
+    "val_iou": 0.7933830893039704,
+    "time_sec": 310.975909948349
+  },
+  {
+    "epoch": 2,
+    "train_loss": 0.3308764501711801,
+    "val_loss": 0.304664246737957,
+    "val_iou": 0.8000004982948303,
+    "time_sec": 287.062824010849
+  },
+  {
+    "epoch": 3,
+    "train_loss": 0.2706743049365218,
+    "val_loss": 0.2952070167660713,
+    "val_iou": 0.7954098653793334,
+    "time_sec": 283.8040874004364
+  },
+  {
+    "epoch": 4,
+    "train_loss": 0.22350290676705725,
+    "val_loss": 0.29507764175534246,
+    "val_iou": 0.7969379591941833,
+    "time_sec": 282.5429263114929
+  },
+  {
+    "epoch": 5,
+    "train_loss": 0.1895671994775854,
+    "val_loss": 0.2825832884758711,
+    "val_iou": 0.8065838325023651,
+    "time_sec": 286.93286657333374
+  },
+  {
+    "epoch": 6,
+    "train_loss": 0.16732191265056637,
+    "val_loss": 0.2762441613525152,
+    "val_iou": 0.8225807750225067,
+    "time_sec": 286.4193227291107
+  },
+  {
+    "epoch": 7,
+    "train_loss": 0.15182167718914674,
+    "val_loss": 0.27510987378656865,
+    "val_iou": 0.8245277297496796,
+    "time_sec": 294.7808041572571
+  },
+  {
+    "epoch": 8,
+    "train_loss": 0.1432231002383762,
+    "val_loss": 0.2713714835047722,
+    "val_iou": 0.8262266063690186,
+    "time_sec": 282.5884385108948
+  }
+]