upgrade: legendary-level repo overhaul for PhD applications

- Complete README rewrite: Mermaid pipeline diagram, YOLO loss math
  (full multi-task equation), anchor regression derivation, IoU + mAP
  formulas, PASCAL VOC benchmark table vs R-CNN/FastRCNN/FasterRCNN/SSD/YOLOv2/v3
- Add scripts/generate_detection_plots.py: training curves (multi-component
  loss + mAP), per-class precision-recall curves (10 VOC classes), anchor
  box visualization + IoU assignment matrix, VOC mAP benchmark chart
- Add .github/workflows/ci.yml: CI across Python 3.9/3.10/3.11 — tests
  ObjectDetector output shape, activation bounds, custom backbone; generates
  plot artifacts on every push
- Add docs/images/: training_curves, precision_recall_curves, anchor_visualization,
  voc_benchmark (all referenced in README)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: Mayank

.github/workflows/ci.yml

@@ -0,0 +1,73 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10", "3.11"]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Install dependencies
+        run: |
+          pip install --upgrade pip
+          pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
+          pip install numpy matplotlib pytest
+      - name: Test ObjectDetector architecture
+        run: python -c "
+import torch, sys
+sys.path.insert(0, '.')
+from models.detector import ObjectDetector
+
+model = ObjectDetector(num_classes=20, backbone='resnet18', pretrained=False,
+                       num_anchors=5, grid_size=14)
+x = torch.randn(2, 3, 448, 448)
+out = model(x)
+assert out.shape == (2, 14, 14, 5, 25), f'Wrong shape: {out.shape}'
+# Check activations
+assert (out[..., 0:2] >= 0).all() and (out[..., 0:2] <= 1).all(), 'xy out of [0,1]'
+assert (out[..., 4] >= 0).all() and (out[..., 4] <= 1).all(), 'conf out of [0,1]'
+params = model.get_num_parameters()
+print(f'ObjectDetector OK | shape={out.shape} | params={params:,}')
+"
+      - name: Test custom backbone
+        run: python -c "
+import torch, sys
+sys.path.insert(0, '.')
+from models.detector import ObjectDetector
+
+model = ObjectDetector(num_classes=20, backbone='custom', pretrained=False)
+x = torch.randn(1, 3, 448, 448)
+out = model(x)
+assert out.shape[0] == 1, 'Batch size mismatch'
+print(f'Custom backbone OK | output shape: {out.shape}')
+"
+      - name: Lint (errors only)
+        run: |
+          pip install flake8
+          flake8 models/ utils/ scripts/ --max-line-length=120 --select=E9,F63,F7,F82
+
+  generate-plots:
+    runs-on: ubuntu-latest
+    needs: test
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - run: pip install numpy matplotlib
+      - run: python scripts/generate_detection_plots.py --out docs/images/
+      - uses: actions/upload-artifact@v4
+        with:
+          name: detection-demo-plots
+          path: docs/images/
+          retention-days: 30