fix: add HF space config to README

README.md

@@ -1,206 +1,11 @@
-# NeuroScreen 🧠
-
-> Vision-based early screening tool for Autism Spectrum Disorder (ASD) and ADHD symptoms.
-> Built for Octopi.Health assignment — Sanjay S, April 2026.
-
----
-
-## What it does
-
-NeuroScreen analyzes images, videos, or live webcam captures and produces a clinical-style screening report covering:
-
-- **ASD risk score** (0–100) based on eye contact, facial expression, social smile, and detected behavioral patterns
-- **ADHD risk score** (0–100) based on head movement, gaze span, blink rate, body fidgeting, and optical flow
-- **Grad-CAM heatmap** showing which facial regions influenced the CNN prediction
-- **Signal breakdown** comparing 8 extracted behavioral signals against clinical reference ranges
-- **Plain-language clinical report** with findings and recommendation
-
-> ⚕️ This is a **screening tool only** — not a medical diagnosis. Always consult a licensed specialist.
-
----
-
-## Architecture
-
-```
-User input (image / video / webcam)
-        ↓
-FastAPI (main.py)
-        ↓
-Orchestrator (analyzer.py)
-    ├── Layer 1: MediaPipe signals (mediapipe_signals.py)
-    │     eye contact, blink rate, expression range,
-    │     head movement, gaze span, fidget score,
-    │     smile score, optical flow
-    ├── Layer 2: CNN classifier (cnn_classifier.py)
-    │     EfficientNetB0 → 9-class softmax → Grad-CAM
-    └── Layer 3: Clinical scoring (scoring.py)
-          ASD score + ADHD score + risk level + report
-        ↓
-JSON report → frontend (index.html)
-```
-
----
-
-## Project structure
-
-```
-neuroscreen/
-├── main.py                    # FastAPI app — all endpoints
-├── app/
-│   ├── __init__.py
-│   ├── analyzer.py            # Orchestration layer
-│   ├── mediapipe_signals.py   # 8 behavioral signal extraction
-│   ├── cnn_classifier.py      # EfficientNetB0 + Grad-CAM
-│   └── scoring.py             # Clinical risk scoring
-├── model/
-│   ├── train.py               # Training script
-│   ├── class_names.json       # 9-class labels
-│   ├── best_model.weights.h5  # Best checkpoint
-│   └── neuroscreen_model.weights.h5  # Final weights
-├── dataset/
-│   ├── train/                 # COCO-format annotations
-│   ├── valid/
-│   └── test/
-├── templates/
-│   └── index.html             # Frontend UI
-├── static/                    # Static assets
-├── uploads/                   # Temp upload dir (auto-cleaned)
-├── requirements.txt
-├── Dockerfile
-└── README.md
-```
-
 ---
-
-## Setup & run locally
-
-### 1. Clone and create virtual environment
-
-```bash
-git clone <your-repo-url>
-cd neuroscreen
-python3 -m venv venv
-source venv/bin/activate
-```
-
-### 2. Install dependencies
-
-```bash
-pip install -r requirements.txt
-```
-
-### 3. Train the model (or use pre-trained weights)
-
-```bash
-python model/train.py
-```
-
-Training uses a proper 70/15/15 stratified split. Expect ~85%+ validation accuracy after both phases.
-
-### 4. Run the server
-
-```bash
-uvicorn main:app --host 0.0.0.0 --port 8000 --reload
-```
-
-Visit: [http://localhost:8000](http://localhost:8000)
-
+title: Neuroscreenn
+emoji: 🧠
+colorFrom: blue
+colorTo: purple
+sdk: docker
+pinned: false
 ---
 
-## API endpoints
-
-| Method | Path | Description |
-|--------|------|-------------|
-| GET | `/` | Frontend UI |
-| POST | `/analyze/image` | 1 or more images (multipart `files`) |
-| POST | `/analyze/video` | Single video file (multipart `file`) |
-| POST | `/analyze/webcam` | JSON `{ "frames": ["base64...", ...] }` |
-| POST | `/analyze/combined` | Images + video together |
-| GET | `/health` | Health check |
-
----
-
-## Model details
-
-| Property | Value |
-|----------|-------|
-| Architecture | EfficientNetB0 + custom head |
-| Input size | 224 × 224 × 3 |
-| Classes | 9 (6 ASD + 1 ADHD + 1 social + 1 normal) |
-| Total params | 4.4M (1.39M trainable) |
-| Training | 2-phase: frozen backbone → fine-tune last 50 layers |
-| Augmentation | Flip, rotation, zoom, brightness, contrast, hue, saturation |
-| Loss | Sparse categorical cross-entropy with label smoothing (0.1) |
-| Regularization | L2 (1e-4) + Dropout (0.5 / 0.4) + class weighting |
-
-### 9 symptom classes
-
-| Class | Condition |
-|-------|-----------|
-| `avoid_eye_contact` | ASD |
-| `hand_flapping` | ASD |
-| `finger_flapping` | ASD |
-| `spinning` | ASD |
-| `stimming` | ASD |
-| `rocking` | ASD |
-| `continuous_moving` | ADHD |
-| `lack_social_skill` | ASD / Social |
-| `normal` | Control |
-
----
-
-## Behavioral signals (MediaPipe layer)
-
-| Signal | Clinical relevance |
-|--------|--------------------|
-| Eye contact duration | Core ASD indicator |
-| Blink rate | ADHD attention marker |
-| Expression range | ASD social communication |
-| Head movement frequency | ADHD hyperactivity |
-| Attention gaze span | ADHD inattention |
-| Body fidgeting | ADHD hyperactivity |
-| Social smile score | ASD emotional reciprocity |
-| Optical flow | ADHD overall movement |
-
----
-
-## Docker
-
-```bash
-docker build -t neuroscreen .
-docker run -p 8000:8000 neuroscreen
-```
-
----
-
-## Requirements
-
-```
-fastapi
-uvicorn[standard]
-tensorflow>=2.13
-mediapipe
-opencv-python-headless
-numpy
-Pillow
-scikit-learn
-matplotlib
-python-multipart
-jinja2
-```
-
----
-
-## Known limitations & future work
-
-- Webcam analysis requires a stable internet connection (frames sent to backend as base64)
-- Model accuracy improves significantly with more diverse training data
-- Currently CPU-only; add GPU support for faster video analysis
-- Future: real-time streaming analysis via WebSockets
-- Future: multi-face detection for classroom scenarios
-
----
-
-## Disclaimer
-
-NeuroScreen is a **research and screening tool**. It is not a certified medical device and should never be used as a substitute for professional clinical assessment by a licensed developmental pediatrician or child psychologist.
+# NeuroScreen
+CNN-based neural screen analyzer using MediaPipe and TensorFlow.