Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+best_phase2.keras filter=lfs diff=lfs merge=lfs -text
+best_phase2_swa.keras filter=lfs diff=lfs merge=lfs -text

best_phase2.keras

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc267da784afaad435bc15a5f01a8e6bcf91c453721b0edfad3254a7c283f355
+size 102521223

best_phase2_swa.keras

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3605c70db3680a8d81be1f305b901a08de5214065a5f9691c7ec1eb7b12d59b2
+size 102521272

config.json

@@ -0,0 +1,117 @@
+{
+  "model_type": "densenet121_custom",
+  "base_model": "tensorflow/densenet121",
+  "task": "image-classification",
+  "num_labels": 6,
+  "id2label": {
+    "0": "bridge",
+    "1": "castle",
+    "2": "mosque",
+    "3": "skyscraper",
+    "4": "stadium",
+    "5": "temple"
+  },
+  "label2id": {
+    "bridge": 0,
+    "castle": 1,
+    "mosque": 2,
+    "skyscraper": 3,
+    "stadium": 4,
+    "temple": 5
+  },
+  "input_shape": [320, 320, 3],
+  "preprocessing": {
+    "mode": "densenet_preprocess_input",
+    "mean": [123.675, 116.28, 103.53],
+    "std": [58.395, 57.12, 57.375],
+    "channel_order": "BGR"
+  },
+  "training_info": {
+    "dataset_source": "Pexels Architecture (10.080 images, 6 classes, balanced)",
+    "dataset_size": 10080,
+    "split_ratio": [0.8, 0.1, 0.1],
+    "split_seed": 42,
+    "phases": [
+      {
+        "name": "Phase 1 - Feature Extraction",
+        "optimizer": "AdamW",
+        "lr": 0.001,
+        "epochs_max": 25,
+        "epochs_actual": 1,
+        "val_accuracy": 0.8671,
+        "val_loss": 1.0042,
+        "train_accuracy": 0.5589,
+        "cutmix_mixup": true,
+        "label_smoothing": 0.1,
+        "early_stop_reason": "myCallback val_acc >= 0.85"
+      },
+      {
+        "name": "Phase 2 - Selective Fine-Tuning",
+        "optimizer": "DiscriminativeAdamW",
+        "lr": 0.0003,
+        "epochs_max": 50,
+        "epochs_actual": 6,
+        "val_accuracy": 0.9335,
+        "val_loss": 0.4877,
+        "train_accuracy": 0.9065,
+        "unfreeze": "conv4_block+conv5_block (BN frozen)",
+        "discriminative_lr": {"conv4_block": 0.1},
+        "cutmix_mixup": false,
+        "label_smoothing": 0.05,
+        "early_stop_reason": "myCallback val_acc >= 0.92"
+      },
+      {
+        "name": "SWA Post-Training",
+        "epochs": 5,
+        "lr": 0.0001,
+        "bn_update": true,
+        "bn_update_steps": 100,
+        "val_accuracy": 0.9593,
+        "val_loss": 0.3981,
+        "method": "Izmailov et al., UAI 2018"
+      }
+    ],
+    "metrics": {
+      "train_accuracy": 0.9945,
+      "val_accuracy": 0.9593,
+      "test_accuracy": 0.9623,
+      "test_loss": 0.3974,
+      "tta_accuracy": 0.9633,
+      "overfitting_gap": 0.0322,
+      "test_correct": 970,
+      "test_total": 1008,
+      "macro_precision": 0.9629,
+      "macro_recall": 0.9623,
+      "macro_f1": 0.9621,
+      "per_class_f1": {
+        "bridge": 0.9529,
+        "castle": 0.9792,
+        "mosque": 0.9593,
+        "skyscraper": 0.9795,
+        "stadium": 0.9412,
+        "temple": 0.9607
+      },
+      "per_class_recall": {
+        "bridge": 0.9643,
+        "castle": 0.9821,
+        "mosque": 0.9821,
+        "skyscraper": 0.9940,
+        "stadium": 0.9048,
+        "temple": 0.9464
+      },
+      "checkpoint_comparison": {
+        "best_phase2_swa": {"val_accuracy": 0.9593, "val_loss": 0.3981, "rank": 1},
+        "best_phase2": {"val_accuracy": 0.9335, "val_loss": 0.4877, "rank": 2},
+        "best_phase2_ema": {"val_accuracy": 0.8720, "val_loss": 0.6621, "rank": 3},
+        "best_phase1": {"val_accuracy": 0.8671, "val_loss": 1.0042, "rank": 4}
+      }
+    }
+  },
+  "version": "v15",
+  "license": "cc-by-4.0",
+  "author": {
+    "name": "Saugani",
+    "email": "mr.asyir@gmail.com"
+  },
+  "dicoding_id": "gani24"
+}

label_mapping.json

@@ -0,0 +1,18 @@
+{
+  "id2label": {
+    "0": "bridge",
+    "1": "castle",
+    "2": "mosque",
+    "3": "skyscraper",
+    "4": "stadium",
+    "5": "temple"
+  },
+  "label2id": {
+    "bridge": 0,
+    "castle": 1,
+    "mosque": 2,
+    "skyscraper": 3,
+    "stadium": 4,
+    "temple": 5
+  }
+}

preprocessor_config.json

@@ -0,0 +1,20 @@
+{
+  "image_processing": {
+    "size": {
+      "width": 320,
+      "height": 320
+    },
+    "resample": "bilinear",
+    "normalize": true,
+    "mode": "densenet_preprocess_input",
+    "channel_order": "RGB",
+    "mean": [123.675, 116.28, 103.53],
+    "std": [58.395, 57.12, 57.375],
+    "scale": 1.0,
+    "description": "Gunakan tf.keras.applications.densenet.preprocess_input untuk preprocessing otomatis. Untuk TF-Lite manual: RGB->BGR, subtract ImageNet mean. Catatan: preprocessor_config.json menyimpan 'BGR' untuk referensi DenseNet ImageNet, tapi densenet.preprocess_input() bekerja pada RGB input dan menangani konversi internal."
+  },
+  "input_name": "input_1",
+  "output_name": "output_1",
+  "input_shape": [1, 320, 320, 3],
+  "output_shape": [1, 6]
+}