fixing gradcam + initializing docs structure

Author: JimWid

docs/challenges.md

@@ -4,14 +4,14 @@ An anomaly-based defect detection system for industrial quality control, built a
 
 ## What This Project Does
 
-This application inspects metal nut images from the [MVTec AD dataset](https://www.mvtec.com/company/research/datasets/mvtec-ad) and classifies them as **Good** or **Defective** using the PatchCore anomaly detection algorithm. It produces heatmaps showing exactly where the model identifies anomalies.
+This application inspects **Steel Surface Defect Classification Dataset**, and classifies them with 4 different types of defect: **defect_1, defect_2, defect_3, defect_4, or no_defect** using PyTorch for a **Convolutional Network architecture** with a **residuals approach**. Plus with a Streamlit app that produces heatmaps showing exactly where the model identifies anomalies.
 
 ## Project Components
 
 | Component | Technology | Purpose |
 |-----------|------------|---------|
 | Frontend | Streamlit | Interactive inspection UI with image display and metrics |
-| Backend | PatchCore (Anomalib) | Anomaly detection model with heatmap generation |
+| Backend | PyTorch CNN Architecture | Anomaly detection model with heatmap generation |
 | Camera Simulation | Custom Python module | Mimics industrial camera acquisition from dataset images |
 | Documentation | MkDocs + Material | This site — project docs and guides |
 | CI/CD | GitHub Actions | Automated testing and linting on every push |
@@ -21,15 +21,15 @@ This application inspects metal nut images from the [MVTec AD dataset](https://w
 
 ```bash
 # Clone and set up
-git clone <your-repo-url>
+git clone JimWid/Anomaly_Detection
 cd anomaly_detection
 
-# Install dependencies (see Installation page for full details)
-pip install -r requirements.txt
+# This version uses UV
+uv sync
 
-# Train the model (one-time, ~2 minutes)
-python -m anomaly_detection.train
+# Train the model (one-time, ~20 minutes)
+python -m steel_defect.train --epochs 20
 
 # Launch the app
-streamlit run anomaly_detection/app.py
+streamlit run final_project/steel_defect/app.py
 ```

docs/data-preprocessing.md

@@ -311,3 +311,35 @@
 2026-08-04 13:18:20 | WARNING  | steel_defect.app          | Grad-CAM failed: 'SteelCNN' object has no attribute 'features'
 2026-08-04 13:18:22 | INFO     | steel_defect.inference    | Inference #68 | label=defect_3 | confidence=0.546 | time=23.3ms
 2026-08-04 13:18:22 | WARNING  | steel_defect.app          | Grad-CAM failed: 'SteelCNN' object has no attribute 'features'
+2026-08-04 13:37:05 | INFO     | steel_defect.inference    | SteelPredictor initialized | checkpoint=C:\Personal Projects\Anomaly_Detection\final_project\models\steel_cnn_best.pt | device=cuda
+2026-08-04 13:37:05 | INFO     | steel_defect.inference    | Model loaded | time=68ms
+2026-08-04 13:37:09 | INFO     | steel_defect.inference    | Inference #1 | label=defect_3 | confidence=0.519 | time=94.1ms
+2026-08-04 13:37:10 | INFO     | steel_defect.inference    | Inference #2 | label=defect_1 | confidence=0.452 | time=2.9ms
+2026-08-04 13:37:12 | INFO     | steel_defect.inference    | Inference #3 | label=defect_1 | confidence=0.520 | time=17.9ms
+2026-08-04 13:37:13 | INFO     | steel_defect.inference    | Inference #4 | label=defect_3 | confidence=0.685 | time=16.3ms
+2026-08-04 13:37:15 | INFO     | steel_defect.inference    | Inference #5 | label=defect_1 | confidence=0.495 | time=16.7ms
+2026-08-04 13:37:17 | INFO     | steel_defect.inference    | Inference #6 | label=defect_3 | confidence=0.544 | time=17.0ms
+2026-08-04 13:37:18 | INFO     | steel_defect.inference    | Inference #7 | label=defect_3 | confidence=0.520 | time=16.5ms
+2026-08-04 13:37:20 | INFO     | steel_defect.inference    | Inference #8 | label=defect_1 | confidence=0.438 | time=16.7ms
+2026-08-04 13:37:21 | INFO     | steel_defect.inference    | Inference #9 | label=defect_3 | confidence=0.627 | time=16.6ms
+2026-08-04 13:37:23 | INFO     | steel_defect.inference    | Inference #10 | label=defect_3 | confidence=0.423 | time=21.1ms
+2026-08-04 13:37:25 | INFO     | steel_defect.inference    | Inference #11 | label=defect_3 | confidence=0.540 | time=21.0ms
+2026-08-04 13:37:26 | INFO     | steel_defect.inference    | Inference #12 | label=defect_3 | confidence=0.456 | time=21.1ms
+2026-08-04 13:37:28 | INFO     | steel_defect.inference    | Inference #13 | label=no_defect | confidence=0.737 | time=21.0ms
+2026-08-04 13:37:29 | INFO     | steel_defect.inference    | Inference #14 | label=defect_1 | confidence=0.443 | time=20.8ms
+2026-08-04 13:37:31 | INFO     | steel_defect.inference    | Inference #15 | label=defect_3 | confidence=0.546 | time=16.4ms
+2026-08-04 13:37:33 | INFO     | steel_defect.inference    | Inference #16 | label=defect_3 | confidence=0.913 | time=17.0ms
+2026-08-04 13:37:34 | INFO     | steel_defect.inference    | Inference #17 | label=defect_1 | confidence=0.456 | time=16.8ms
+2026-08-04 13:37:36 | INFO     | steel_defect.inference    | Inference #18 | label=defect_1 | confidence=0.512 | time=16.4ms
+2026-08-04 13:37:37 | INFO     | steel_defect.inference    | Inference #19 | label=no_defect | confidence=0.540 | time=16.4ms
+2026-08-04 13:37:39 | INFO     | steel_defect.inference    | Inference #20 | label=defect_1 | confidence=0.445 | time=16.7ms
+2026-08-04 13:37:41 | INFO     | steel_defect.inference    | Inference #21 | label=defect_3 | confidence=0.589 | time=20.9ms
+2026-08-04 13:37:42 | INFO     | steel_defect.inference    | Inference #22 | label=defect_3 | confidence=0.550 | time=21.1ms
+2026-08-04 13:37:45 | INFO     | steel_defect.inference    | Inference #23 | label=no_defect | confidence=0.720 | time=23.5ms
+2026-08-04 13:37:47 | INFO     | steel_defect.inference    | Inference #24 | label=no_defect | confidence=0.874 | time=20.7ms
+2026-08-04 13:37:48 | INFO     | steel_defect.inference    | Inference #25 | label=no_defect | confidence=1.000 | time=20.9ms
+2026-08-04 13:37:50 | INFO     | steel_defect.inference    | Inference #26 | label=defect_3 | confidence=0.533 | time=20.8ms
+2026-08-04 13:37:52 | INFO     | steel_defect.inference    | Inference #27 | label=no_defect | confidence=0.544 | time=20.9ms
+2026-08-04 13:37:53 | INFO     | steel_defect.inference    | Inference #28 | label=no_defect | confidence=0.992 | time=20.7ms
+2026-08-04 13:37:55 | INFO     | steel_defect.inference    | Inference #29 | label=no_defect | confidence=0.866 | time=16.7ms
+2026-08-04 13:37:56 | INFO     | steel_defect.inference    | Inference #30 | label=no_defect | confidence=0.705 | time=16.6ms

docs/index.md

@@ -78,9 +78,9 @@ def get_images_in_category(category: str) -> list[Path]:
 def generate_gradcam(predictor: SteelPredictor, image_tensor: torch.Tensor) -> np.ndarray:
     """Generate Grad-CAM heatmap from the last conv layer."""
     try:
-        # Get the last convolutional layer from model.features
+        # Get the last convolutional layer from model.features / this verson uses 'residual_block_3' as the last block
         target_layer = None
-        for layer in reversed(list(predictor.model.features.children())):
+        for layer in reversed(list(predictor.model.residual_block_3.children())):
             if isinstance(layer, torch.nn.Conv2d):
                 target_layer = layer
                 break

docs/model-architecture.md

@@ -13,11 +13,15 @@ theme:
 
 nav:
   - Home: index.md
+  - Architecture: model-architecture.md
+  - Data Preprocessing: data-preprocessing.md
+  - Training: training.md
+  - Results: results.md
+  - Challenges: challenges.md
   - Installation: installation.md
   - Usage Guide: usage.md
-  - Architecture: architecture.md
-  - Advanced - Image Acquisition: advanced-acquisition.md
-  - Homework:
+
+  - Homework - DONE:
     - Overview: homework/index.md
     - Step 1 - Preprocessing: homework/step-1-preprocessing.md
     - Step 2 - Dataset: homework/step-2-dataset.md