Upload V16 README.md

README.md

@@ -8,13 +8,15 @@ tags:
   - densenet121
   - architecture
   - building
-  - cnn
   - fgvc
   - transfer-learning
   - gem-pooling
   - focal-loss
-  - swa
   - discriminative-learning-rate
+  - swa
+  - grad-cam
+  - calibration
+  - roc-auc
 library_name: keras
 language: en
 datasets:
@@ -34,41 +36,45 @@ model-index:
           split: test
         metrics:
           - type: accuracy
-            value: 96.23
+            value: 0.9688
             name: Test Accuracy
           - type: accuracy
-            value: 95.93
+            value: 0.9658
             name: Validation Accuracy (SWA)
           - type: accuracy
-            value: 96.33
+            value: 0.968
             name: TTA Accuracy
 ---
 
-# Architectural Building Image Classifier
+# Fine-Grained World Architecture Image Classification: A DenseNet121 Transfer Learning Approach with Layered Regularization
 
-Fine-Grained Visual Categorization (FGVC) of world architectural buildings using CNN transfer learning with DenseNet121, enhanced with GeM Pooling, Focal Loss, and Stochastic Weight Averaging (SWA).
+### Architectural Building Image Classifier
+
+Fine-Grained Visual Classification (FGVC) of world architectural buildings using CNN transfer learning with DenseNet121, enhanced with GeM Pooling, Focal Loss, Discriminative AdamW (LR), Stochastic Weight Averaging (SWA), Grad-CAM explainability, and calibration analysis.
 
 <table>
-  <tr><td><strong>Architecture</strong></td><td>DenseNet121 + GeM Pooling (p=3.0) + SWA</td></tr>
-  <tr><td><strong>Task</strong></td><td>Fine-Grained Visual Categorization (FGVC)</td></tr>
-  <tr><td><strong>Test Accuracy</strong></td><td>96.23% (970/1,008)</td></tr>
-  <tr><td><strong>Classes</strong></td><td>6 (Bridge, Castle, Mosque, Skyscraper, Stadium, Temple)</td></tr>
+  <tr><td><strong>Architecture</strong></td><td>DenseNet121 + GeM Pooling + Focal Loss + SWA</td></tr>
+  <tr><td><strong>Task</strong></td><td>Fine-Grained Visual Classification (FGVC)</td></tr>
+  <tr><td><strong>Test Accuracy</strong></td><td>96.88%</td></tr>
+  <tr><td><strong>Classes</strong></td><td>8 (Barn, Bridge, Castle, Mosque, Skyscraper, Stadium, Temple, Windmill)</td></tr>
   <tr><td><strong>Input Size</strong></td><td>320 × 320 pixels</td></tr>
-  <tr><td><strong>Parameters</strong></td><td>9,466,439 (9.27M trainable in Phase 2)</td></tr>
+  <tr><td><strong>Parameters</strong></td><td>9,466,953</td></tr>
   <tr><td><strong>Framework</strong></td><td>TensorFlow / Keras 3</td></tr>
   <tr><td><strong>License</strong></td><td><a href="https://www.apache.org/licenses/LICENSE-2.0">Apache-2.0</a></td></tr>
 </table>
 
 ## Model Description
 
-A fine-grained image classification model for world architectural buildings. Built on DenseNet121 pretrained on ImageNet, enhanced with GeM Pooling (learnable generalized mean pooling), Focal Loss, and Stochastic Weight Averaging (SWA).
+A fine-grained image classification model for world architectural buildings. Built on DenseNet121 pretrained on ImageNet, enhanced with GeM Pooling (learnable generalized mean pooling), Focal Loss, Discriminative AdamW and Stochastic Weight Averaging (SWA). Extended with Grad-CAM explainability visualization, ROC-AUC evaluation, ECE calibration analysis, and t-SNE embedding visualization.
 
 **Key architectural contributions:**
 
 - **GeM Pooling** (Radenovic et al., CVPR 2018) — replaces global average pooling with a learnable power parameter (p=3.0) that emphasizes high-activation features, yielding stronger discriminative representations for FGVC tasks
 - **Focal Loss** (Lin et al., ICCV 2017, gamma=2.0) — down-weights well-classified examples to focus gradient updates on hard-to-classify building pairs
 - **DiscriminativeAdamW** — extends AdamW with per-layer learning rate multipliers: conv4_block receives LR × 0.1 (pretrained features require smaller updates), while conv5_block and the custom head receive LR × 1.0
-- **SWA with BN re-estimation** (Izmailov et al., UAI 2018) — 5-epoch post-training weight averaging with constant LR 1e-4, followed by 100-step batch normalization statistics re-estimation, improving validation accuracy from 93.35% to 95.93%
+- **SWA with BN re-estimation** (Izmailov et al., UAI 2018) — 10-epoch post-training weight averaging with constant LR 1e-4, followed by 100-step batch normalization statistics re-estimation
+- **Grad-CAM** (Selvaraju et al., ICCV 2017) — gradient-weighted class activation mapping for explainability, targeting *conv5_block16_concat*
+- **ECE Calibration** (Guo et al., ICML 2017) — Expected Calibration Error with 15-bin reliability diagram to assess prediction confidence reliability
 
 ## Architecture
 
@@ -87,9 +93,9 @@ Input (320, 320, 3)
   BatchNormalization                        →  1,024 params
   Dropout(0.4)                              →  0 params
   │
-  Dense(6, Softmax)                         →  1,542 params
+  Dense(8, Softmax)                         →  2,056 params
   │
-Output (6 classes)
+Output (8 classes)
 ```
 
 | Component | Output Shape | Parameters |
@@ -102,10 +108,10 @@ Output (6 classes)
 | Dense 256 ReLU | (None, 256) | 65,792 |
 | BatchNormalization | (None, 256) | 1,024 |
 | Dropout 0.4 | (None, 256) | 0 |
-| Dense 6 Softmax | (None, 6) | 1,542 |
-| **Total** | | **9,466,439** |
-| Trainable (Phase 1) | | **2,427,911** (9.26 MB) |
-| Trainable (Phase 2) | | **7,883,783** (30.07 MB) |
+| Dense 8 Softmax | (None, 8) | 2,056 |
+| **Total** | | **9,466,953** |
+| Trainable (Phase 1) | | **2,428,425** (9.27 MB) |
+| Trainable (Phase 2) | | **7,884,297** (30.07 MB) |
 | Non-trainable (Phase 1) | | **7,038,528** (26.85 MB) |
 
 ## Performance
@@ -114,29 +120,32 @@ Output (6 classes)
 
 | Metric | Value |
 |--------|-------|
-| Test Accuracy | 96.23% (970/1,008) |
-| Validation Accuracy (SWA) | 95.93% |
-| Test-Time Augmentation | 96.33% (+0.10%) |
-| Test Loss | 0.3974 |
-| Overfitting Gap (Train − Test) | 3.22% |
-| Macro Avg Precision | 96.29% |
-| Macro Avg Recall | 96.23% |
-| Macro Avg F1-Score | 96.21% |
+| Test Accuracy | 96.88% |
+| Validation Accuracy (SWA) | 96.58% |
+| Test-Time Augmentation | 96.80% |
+| Test Loss | 0.4485 |
+| Overfitting Gap (Train − Test) | 3.00% |
+| Macro Avg Precision | 0.9691 |
+| Macro Avg Recall | 0.9688 |
+| Macro Avg F1-Score | 0.9687 |
+| Top-2 Accuracy | 98.59% |
+| Top-3 Accuracy | 99.33% |
+| Macro ROC-AUC (OvR) | 0.9986 |
+| ECE (15 bins) | 0.1438 |
 
 ### Per-Class Results
 
 | Class | Precision | Recall | F1-Score | Support |
 |-------|-----------|--------|----------|---------|
-| Bridge | 94.19% | 96.43% | 95.29% | 168 |
-| Castle | 97.63% | 98.21% | 97.92% | 168 |
-| Mosque | 93.75% | 98.21% | 95.93% | 168 |
-| Skyscraper | 96.53% | 99.40% | 97.95% | 168 |
-| Stadium | 98.06% | 90.48% | 94.12% | 168 |
-| Temple | 97.55% | 94.64% | 96.07% | 168 |
-| **Macro Avg** | **96.29%** | **96.23%** | **96.21%** | **1,008** |
-
-**Highest performing classes:** Skyscraper (recall=99.40%), Castle (F1=97.92%)  
-**Most challenging classes:** Stadium (recall=90.48%), Bridge (precision=94.19%)
+| Barn | 0.9645 | 0.9702 | 0.9674 | 168 |
+| Bridge | 0.9588 | 0.9702 | 0.9645 | 168 |
+| Castle | 0.9649 | 0.9821 | 0.9735 | 168 |
+| Mosque | 0.9649 | 0.9821 | 0.9735 | 168 |
+| Skyscraper | 0.9708 | 0.9881 | 0.9794 | 168 |
+| Stadium | 0.9936 | 0.9286 | 0.9600 | 168 |
+| Temple | 0.9816 | 0.9524 | 0.9668 | 168 |
+| Windmill | 0.9535 | 0.9762 | 0.9647 | 168 |
+| **Macro Avg** | **0.9691** | **0.9688** | **0.9687** | **1,344** |
 
 ### Model Selection
 
@@ -144,27 +153,42 @@ Four candidate models were evaluated on the validation set:
 
 | Checkpoint | Val Accuracy | Val Loss | Description |
 |------------|-------------|----------|-------------|
-| `best_phase1.keras` | 86.71% | 1.0042 | Phase 1 checkpoint (backbone frozen) |
-| `best_phase2.keras` | 93.35% | 0.4877 | Phase 2 checkpoint (conv4+conv5 unfrozen) |
-| `best_phase2_ema.keras` | 87.20% | 0.6621 | Phase 2 EMA shadow weights |
-| **`best_phase2_swa.keras`** | **95.93%** | **0.3981** | **SWA averaged weights ← SELECTED** |
+| `best_phase1.keras` | 89.21% | 1.2231 | Phase 1 checkpoint (backbone frozen) |
+| `best_phase2.keras` | 92.04% | 0.6171 | Phase 2 checkpoint (conv4+conv5 unfrozen) |
+| `best_phase2_ema.keras` | 89.36% | 0.8183 | Phase 2 EMA shadow weights |
+| **`best_phase2_swa.keras`** | **96.58%** | **0.4256** | **SWA averaged weights ← SELECTED** |
 
 ### SWA Progression
 
 | SWA Epoch | Val Accuracy | Val Loss |
 |-----------|-------------|----------|
-| 1 | 92.56% | 0.5182 |
-| 2 | 94.74% | 0.4691 |
-| 3 | 93.75% | 0.4694 |
-| 4 | 95.93% | 0.4297 |
-| 5 | 95.83% | 0.4130 |
-| **SWA Average (final)** | **95.93%** | **0.3981** |
+| 1 | 93.38% | 0.5580 |
+| 2 | 93.60% | 0.5738 |
+| 3 | 92.86% | 0.5725 |
+| 4 | 95.24% | 0.4806 |
+| 5 | 95.68% | 0.4529 |
+| 6 | 96.35% | 0.4548 |
+| 7 | 94.27% | 0.5141 |
+| 8 | 94.12% | 0.5147 |
+| 9 | 94.49% | 0.5243 |
+| 10 | 96.50% | 0.4424 |
+| **SWA + BN (final)** | **96.58%** | **0.4256** |
+
+![Training Curves](results/training_curves.png)
+
+![Confusion Matrix](results/confusion_matrix.png)
+
+![Per-Class Accuracy](results/per_class_accuracy.png)
 
-![Training Curves](training_curves.png)
+![Reliability Diagram](results/reliability_diagram.png)
 
-![Confusion Matrix](confusion_matrix.png)
+![ROC Curves](results/roc_curves.png)
 
-![Per-Class Accuracy](per_class_accuracy.png)
+![t-SNE Embedding](results/tsne_embedding.png)
+
+![Grad-CAM Heatmaps](results/gradcam_heatmaps.png)
+
+![Confidence Per Class](results/confidence_per_class.png)
 
 ## Training Details
 
@@ -174,12 +198,12 @@ Two-phase progressive training with SWA post-processing:
 
 | Phase | Description | Backbone | Optimizer | LR | Max Epochs | Actual Epochs | CutMix+Mixup | FocalLoss LS |
 |-------|-------------|----------|-----------|-----|-----------|---------------|---------------|-------------|
-| **Phase 1** — Feature Extraction | Train custom head only | Frozen (all) | AdamW (wd=2e-5) | 0.001 + CosineDecay + Warmup 3ep | 25 | 1 ¹ | Yes (50/50 alternation) | 0.1 |
-| **Phase 2** — Selective Fine-Tuning | Load best_phase1 → fine-tune | conv4_block + conv5_block unfrozen (BN frozen) | DiscriminativeAdamW (conv4=0.1×) | 3e-4 + CosineDecay + Warmup 5ep | 50 | 6 + 5 SWA ² | No | 0.05 |
+| **Phase 1** — Feature Extraction | Train custom head only | Frozen (all) | AdamW (wd=2e-5) | 0.001 + CosineDecay + Warmup 3ep | 25 | 1 | Yes (50/50 alternation) | 0.1 |
+| **Phase 2** — Selective Fine-Tuning | Load best_phase1 → fine-tune | conv4_block + conv5_block unfrozen (BN frozen) | DiscriminativeAdamW (conv4=0.1×) | 3e-4 + CosineDecay + Warmup 5ep | 50 | 6 + 10 SWA | No | 0.05 |
 
-> ¹ Phase 1 stopped at epoch 1 because `val_accuracy = 86.71% ≥ 85%` threshold (myCallback). This demonstrates the effectiveness of ImageNet transfer learning — a single epoch of head training exceeds the target.
+> ¹ Phase 1 stops when `val_accuracy ≥ 85%` threshold (myCallback).
 
-> ² Phase 2 stopped at epoch 6 with `val_accuracy = 93.35%` (≥ 92% threshold), followed by 5 SWA epochs (constant LR 1e-4) improving to 95.93%.
+> ² Phase 2 stops when `val_accuracy ≥ 92%` threshold (myCallback), followed by 10 SWA epochs (constant LR 1e-4).
 
 ### Hyperparameters
 
@@ -196,7 +220,7 @@ Two-phase progressive training with SWA post-processing:
 | Early Stopping Patience | 7 | 12 |
 | myCallback Threshold | val_acc ≥ 0.85 | val_acc ≥ 0.92 |
 | EMA Decay | 0.999 | 0.999 |
-| SWA Epochs | — | 5 (post-training) |
+| SWA Epochs | — | 10 (post-training) |
 | SWA LR | — | 1×10⁻⁴ (constant) |
 | BN Re-estimation Steps | — | 100 |
 | CutMix (alpha=1.0) | Yes (50% batches) | No |
@@ -217,7 +241,7 @@ Two-phase progressive training with SWA post-processing:
 | Dropout | 0.4 after Dense(256)+BN | Srivastava et al., JMLR 2014 |
 | Batch Normalization | After Conv2D and Dense; frozen during fine-tuning | Ioffe & Szegedy, arXiv 2015 |
 | EMA | Shadow weights, decay=0.999 | Tarvainen & Valpola, NeurIPS 2017 |
-| SWA | 5-epoch post-training, constant LR 1e-4 | Izmailov et al., UAI 2018 |
+| SWA | 10-epoch post-training, constant LR 1e-4 | Izmailov et al., UAI 2018 |
 | Data Augmentation | Rotation ±15°, shift ±10%, zoom ±20%, brightness 0.75–1.15, horizontal flip | Perez & Wang, arXiv 2017 |
 | Test-Time Augmentation | 6 augmentation variants, averaged | Shanmugam et al., ICML 2020 |
 | WarmupCosineDecay | Linear warmup + cosine annealing | Loshchilov & Hutter, ICLR 2017 (SGDR) |
@@ -225,13 +249,13 @@ Two-phase progressive training with SWA post-processing:
 
 ### Dataset
 
-[0xgr3y/arch-building-dataset](https://huggingface.co/datasets/0xgr3y/arch-building-dataset) — 10,080 images (6 classes × 1,680, balanced) sourced from Pexels with perceptual (pHash) and exact (SHA256) deduplication.
+[0xgr3y/arch-building-dataset](https://huggingface.co/datasets/0xgr3y/arch-building-dataset) — 13,440 images (8 classes × 1,680, balanced) sourced from Pexels with perceptual (pHash) and exact (SHA256) deduplication.
 
 | Split | Images | Percentage |
 |-------|--------|------------|
-| Train | 8,064 | 80% |
-| Validation | 1,008 | 10% |
-| Test | 1,008 | 10% |
+| Train | 10,752 | 80% |
+| Validation | 1,344 | 10% |
+| Test | 1,344 | 10% |
 
 ### Data Preprocessing
 
@@ -244,21 +268,37 @@ Two-phase progressive training with SWA post-processing:
 
 | File | Description |
 |------|-------------|
-| `best_phase2_swa.keras` | Best model — SWA averaged weights (val_acc=95.93%) |
-| `best_phase2.keras` | Phase 2 checkpoint (val_acc=93.35%) |
+| `best_phase2_swa.keras` | Best model — SWA averaged weights |
+| `best_phase2.keras` | Phase 2 checkpoint |
 | `saved_model/` | TensorFlow SavedModel format (portable, for TF Serving) |
 | `tflite/model.tflite` | TensorFlow Lite model (mobile/embedded) |
 | `tflite/label.txt` | Class label names for TF-Lite |
 | `tfjs_model/` | TensorFlow.js model (browser, 10 weight shards + model.json) |
 | `config.json` | Model configuration and evaluation metrics |
-| `label_mapping.json` | Class name ↔ ID mapping (integer values) |
+| `label_mapping.json` | Class name ↔ ID mapping with training config and architecture info |
 | `preprocessor_config.json` | Input preprocessing specification (320×320) |
+| `confusion_pairs.json` | Auto-detected confusion pairs from confusion matrix (threshold >5%) |
+| `class_confidence_stats.json` | Per-class mean/std/p5/p95 confidence distribution |
+| `model_benchmark.json` | Model parameters, sizes, speed, Top-K, AUC, ECE, TTA metrics |
+| `calibration_data.json` | ECE, bin accuracies/confidences, per-class AUC for calibration analysis |
+| `results/training_curves.png` | Training/validation accuracy and loss curves |
+| `results/confusion_matrix.png` | Confusion matrix on test set |
+| `results/per_class_accuracy.png` | Per-class accuracy bar chart |
+| `results/reliability_diagram.png` | ECE calibration reliability diagram (15 bins) |
+| `results/roc_curves.png` | Per-class ROC curves (One-vs-Rest) |
+| `results/tsne_embedding.png` | t-SNE 2D scatter plot from GeM Pooling features |
+| `results/gradcam_heatmaps.png` | Grad-CAM heatmap visualization per class |
+| `results/confidence_per_class.png` | Per-class confidence distribution bar chart |
+| `results/misclassification_examples.png` | Misclassified test samples |
+| `results/augmentation_examples.png` | Example images after augmentation (training set) |
+| `results/inference_keras.png` | Keras inference grid — 1 sample per class |
+| `results/inference_tflite.png` | TF-Lite inference grid — 1 sample per class |
 
 ## Usage
 
-### Gradio Demo
+### Gradio Space
 
-Try the live demo: [arch-building-classifier Space](https://huggingface.co/spaces/0xgr3y/arch-building-classifier)
+Try the live inference: [arch-building-classifier Space](https://huggingface.co/spaces/0xgr3y/arch-building-classifier)
 
 ### Python — Keras
 
@@ -282,7 +322,7 @@ class GeMPooling(Layer):
             initializer=tf.keras.initializers.Constant(self.p_init), trainable=True)
         super().build(input_shape)
     def call(self, x):
-        x = tf.clip_by_value(x, self.eps, tf.reduce_max(x))
+        x = tf.maximum(x, self.eps)
         x = tf.pow(x, self.p)
         x = tf.reduce_mean(x, axis=[1, 2], keepdims=False)
         return tf.pow(x, 1.0 / self.p)
@@ -347,7 +387,12 @@ class DiscriminativeAdamW(tf.keras.optimizers.AdamW):
 
 # =====================---Load Model---==========================
 
-LABELS = ["bridge", "castle", "mosque", "skyscraper", "stadium", "temple"]
+LABELS_PATH = hf_hub_download("0xgr3y/Arch-Building-Image-Classification", "label_mapping.json")
+import json
+with open(LABELS_PATH) as f:
+    LABELS = json.load(f)["labels"]
+# ["barn","bridge","castle","mosque","skyscraper","stadium","temple","windmill"]
+
 custom_objects = {
     "GeMPooling": GeMPooling,
     "FocalLoss": FocalLoss,
@@ -359,7 +404,7 @@ model = tf.keras.models.load_model(model_path, custom_objects=custom_objects, co
 
 # =======================---Inference---==========================
 
-img = Image.open("building.jpg").convert("RGB").resize((320, 320))
+img = Image.open("skyscraper_00000.jpg").convert("RGB").resize((320, 320))
 arr = np.expand_dims(preprocess_input(np.array(img, dtype=np.float32)), axis=0)
 preds = model.predict(arr, verbose=0)[0]
 print(f"Predicted: {LABELS[np.argmax(preds)]} ({np.max(preds)*100:.1f}%)")
@@ -377,9 +422,13 @@ hf_hub_download("0xgr3y/Arch-Building-Image-Classification", "tflite/label.txt",
 ```python
 import numpy as np
 import tensorflow as tf
+from huggingface_hub import hf_hub_download
 from PIL import Image
+import json
 
-LABELS = ["bridge", "castle", "mosque", "skyscraper", "stadium", "temple"]
+labels_path = hf_hub_download("0xgr3y/Arch-Building-Image-Classification", "label_mapping.json")
+with open(labels_path) as f:
+    LABELS = json.load(f)["labels"]
 
 interpreter = tf.lite.Interpreter(model_path="tflite/model.tflite")
 interpreter.allocate_tensors()
@@ -387,7 +436,7 @@ interpreter.allocate_tensors()
 input_details = interpreter.get_input_details()
 output_details = interpreter.get_output_details()
 
-img = Image.open("building.jpg").convert("RGB").resize((320, 320))
+img = Image.open("skyscraper_00000.jpg").convert("RGB").resize((320, 320))
 arr = np.expand_dims(np.array(img, dtype=np.float32), axis=0)
 arr = tf.keras.applications.densenet.preprocess_input(arr)
 
@@ -413,15 +462,20 @@ Requires the custom layer definitions from the Keras section above.
 ```python
 import tensorflow as tf
 import numpy as np
+from huggingface_hub import hf_hub_download
 from PIL import Image
+import json
+
+labels_path = hf_hub_download("0xgr3y/Arch-Building-Image-Classification", "label_mapping.json")
+with open(labels_path) as f:
+    LABELS = json.load(f)["labels"]
 
-LABELS = ["bridge", "castle", "mosque", "skyscraper", "stadium", "temple"]
 custom_objects = {"GeMPooling": GeMPooling, "FocalLoss": FocalLoss,
                   "DiscriminativeAdamW": DiscriminativeAdamW}
 
 model = tf.keras.models.load_model("saved_model", custom_objects=custom_objects, compile=False)
 
-img = Image.open("building.jpg").convert("RGB").resize((320, 320))
+img = Image.open("skyscraper_00000.jpg").convert("RGB").resize((320, 320))
 arr = np.expand_dims(tf.keras.applications.densenet.preprocess_input(
     np.array(img, dtype=np.float32)), axis=0)
 preds = model.predict(arr, verbose=0)[0]
@@ -432,31 +486,30 @@ print(f"Predicted: {LABELS[np.argmax(preds)]} ({np.max(preds)*100:.1f}%)")
 
 - Architectural style classification from building photographs
 - Educational tool for architecture recognition
-- Research baseline for fine-grained visual categorization (FGVC)
+- Research baseline for fine-grained visual classification (FGVC)
 - Transfer learning experiments on architectural imagery
 
 ## Limitations
 
 - Trained on Pexels stock photography — performance may differ on user-generated or field photographs
-- Limited to 6 architectural classes (bridge, castle, mosque, skyscraper, stadium, temple)
-- **Stadium** has the lowest recall (90.48%) — often confused with temple due to similar structural outlines
-- **Temple** is the most frequently misclassified class — confused with mosque (shared domes, minarets, and ornamental features)
-- **Skyscraper↔Bridge** confusion occurs when bridge structural elements resemble tall buildings against skyline
-- Inference confidence can be low on atypical examples (e.g., skyscraper predicted at 57.0% confidence with bridge at 36.2%)
+- Limited to 8 architectural classes (barn, bridge, castle, mosque, skyscraper, stadium, temple, windmill)
+- Confusion pair analysis found **0 significant pairs** (threshold >5%) — all 8 classes are well-distinguished by the model; see `confusion_pairs.json` for details
+- Barn and windmill share 3 cross-class duplicates (0.02% of dataset) — left as-is due to negligible impact
+- Inference confidence can be low on atypical examples
 
-![Misclassification Examples](misclassification_examples.png)
+![Misclassification Examples](results/misclassification_examples.png)
 
 ## Ethical Considerations
 
 - All training images sourced from [Pexels.com](https://www.pexels.com) under the Pexels License (free for commercial use, no attribution required). No copyrighted or personally identifiable images were used.
 - The dataset contains only photographs of buildings and structures — no people, faces, or private property are the subject of classification.
 - The model reflects the visual distribution of Pexels stock photography, which may over-represent Western and iconic architectural styles and under-represent vernacular or regional architecture.
-- The 6 class categories are broad and do not capture the full diversity of world architecture. Results should not be used to make definitive claims about architectural categorization.
+- The 8 class categories are broad and do not capture the full diversity of world architecture. Results should not be used to make definitive claims about architectural categorization.
 - URL pattern filtering during dataset collection explicitly excluded AI-generated art, illustrations, and non-photographic content to ensure authenticity.
 
 ## Links
 
-- **Gradio Demo:** [arch-building-classifier Space](https://huggingface.co/spaces/0xgr3y/arch-building-classifier)
+- **Gradio Space (Live):** [arch-building-classifier Space](https://huggingface.co/spaces/0xgr3y/arch-building-classifier)
 - **Dataset:** [0xgr3y/arch-building-dataset](https://huggingface.co/datasets/0xgr3y/arch-building-dataset)
 - **GitHub:** [arcxteam/arch-building-classifier](https://github.com/arcxteam/arch-building-classifier)
 
@@ -478,14 +531,19 @@ print(f"Predicted: {LABELS[np.argmax(preds)]} ({np.max(preds)*100:.1f}%)")
 14. Shanmugam, D., Blalock, D., Balakrishnan, G., Guttag, J., & Sarma, A. (2020). Towards Principled Test-Time Augmentation. *ICML 2020*. [PDF](https://dmshanmugam.github.io/pdfs/icml_2020_testaug.pdf)
 15. Loshchilov, I., & Hutter, F. (2017). SGDR: Stochastic Gradient Descent with Warm Restarts. *ICLR 2017*. [arXiv:1608.03983](https://arxiv.org/abs/1608.03983)
 16. Prechelt, L. (1998). Automatic Early Stopping Using Cross Validation: Quantifying the Criteria. *Neural Networks*, 11(4), 761–767. [https://doi.org/10.1016/S0893-6080(98)00010-0](https://doi.org/10.1016/S0893-6080(98)00010-0)
+17. Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. (2017). On Calibration of Modern Neural Networks. *ICML 2017*. [arXiv:1706.04599](https://arxiv.org/abs/1706.04599)
+18. Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2017). Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization. *ICCV 2017*. [arXiv:1610.02391](https://arxiv.org/abs/1610.02391)
+19. van der Maaten, L., & Hinton, G. (2008). Visualizing Data using t-SNE. *JMLR*, 9(Nov), 2579–2605. [http://jmlr.org/papers/v9/vandermaaten08a.html](http://jmlr.org/papers/v9/vandermaaten08a.html)
+20. Hand, D. J., & Till, R. J. (2001). A Simple Generalisation of the Area Under the ROC Curve for Multiple Class Classification Problems. *Machine Learning*, 45(2), 171–186. [https://doi.org/10.1023/A:1010920819831](https://doi.org/10.1023/A:1010920819831)
+21. Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., ... & Fei-Fei, L. (2015). ImageNet Large Scale Visual Recognition Challenge. *IJCV*, 115(3), 211–252. [arXiv:1409.0575](https://arxiv.org/abs/1409.0575)
+22. Lakshminarayanan, B., Pritzel, A., & Blundell, C. (2017). Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles. *NeurIPS 2017*. [arXiv:1612.01474](https://arxiv.org/abs/1612.01474)
 
 ## Citation
 
 ```bibtex
 @misc{saugani2026_arch_building,
-  title={Fine-Grained Visual Categorization (FGVC) of World Architectural Buildings
-         Using CNN Transfer Learning DenseNet121 with Fine-Tuning and
-         Multi-Layer Regularization Strategy},
+  title={Fine-Grained World Architecture Image Classification:
+         A DenseNet121 Transfer Learning Approach with Layered Regularization},
   author={Saugani},
   year={2026},
   publisher={Hugging Face},