Upload README.md with huggingface_hub

README.md

@@ -3,55 +3,61 @@ language:
 - en
 license: mit
 tags:
-- image-classification
-- pytorch
-- convnext
-- plant-disease
 - agriculture
+- plant-pathology
+- convnext
+- deep-learning
 - edge-cloud
+metrics:
+- accuracy
 pipeline_tag: image-classification
-model-index:
-- name: ConvNeXt-Large Cloud Plant Classifier
-  results:
-  - task:
-      type: image-classification
-      name: Image Classification
-    dataset:
-      name: Plant Disease Classification Merged Dataset (88 Classes)
-      type: plant-disease-classification-merged-dataset
-    metrics:
-    - type: accuracy
-      value: 96.42
-      name: Validation Accuracy
 ---
 
-# ConvNeXt-Large Cloud Plant Classifier
+# ConvNeXt-Large Cloud Plant Disease Diagnostician
+
+This repository hosts the high-capacity **Cloud Classifier (ConvNeXt-Large)** for the *Adaptive Edge-Cloud Plant Disease Diagnosis* framework. The model is dynamically queried by edge nodes running `MobileNetV4-EDL` when classification confidence is low or uncertainty is high.
 
-This repository hosts the **ConvNeXt-Large** cloud classifier model trained as part of the *Confidence-Aware Adaptive Edge–Cloud Framework for Reliable Plant Disease Diagnosis in Low-Connectivity Agricultural Environments*.
+## 1. Mathematical and Framework Documentation
+A complete mathematical report detailing the Evidential Deep Learning (EDL), Unsupervised Domain Adaptation (UDA), and Conformal Calibration models is compiled as a PDF and available in this repository:
+👉 **[Read the Mathematical Report (PDF)](https://huggingface.co/Arko007/adaptive-cloud-plant-model/blob/main/model_report.pdf)**
 
-The cloud model serves as the high-capacity, heavy diagnostician tier of the cooperative inference pipeline. It is triggered dynamically by the edge node when a captured leaf image exhibits high epistemic uncertainty (vacuity u > tau_vac) or low conformal confidence (p_max < tau_conf).
+---
+
+## 2. Model Architecture and Training Details
 
-## Model Architecture
-- **Backbone**: ConvNeXt-Large (originally pre-trained on ImageNet-1K)
-- **Classifier Head**: Modified Linear Layer projecting to 88 pathological crop classes
-- **Input Dimension**: 384 x 384 pixels (higher resolution to capture fine-grained leaf spots)
+- **Model Type**: ConvNeXt-Large
+- **Number of Classes**: 88 (spanning various crop types including Apple, Tomato, Wheat, Soybean, Sugarcane, Tea, etc.)
+- **Resolution**: $384 	imes 384$ pixels
+- **Optimization Strategy**: Stages 1-3 of the backbone were frozen to optimize GPU efficiency.
+- **Optimizer**: AdamW (Learning Rate: $10^{-4}$, Weight Decay: $10^{-3}$)
+- **Loss Function**: Categorical Cross-Entropy
 
-## Training Setup & Parameters
-- **Dataset**: Merged Plant Disease Dataset (88 unique classes spanning laboratory and field-captured folders).
-- **Optimization Strategy**: Parameter-Efficient Fine-Tuning (PEFT) with convolutional stages 1–3 frozen, training only the 4th stage and classifier head.
-- **Optimizer**: AdamW (Learning Rate: 1e-4, Weight Decay: 1e-3)
-- **Loss Function**: Cross-Entropy Loss
-- **Train/Val Split**: 85% Train / 15% Validation (Stratified per-folder split)
+### Training & Validation Loss Curves
+![Loss Curve](https://huggingface.co/Arko007/adaptive-cloud-plant-model/resolve/main/loss_curve.png)
 
-## Operational Performance (Epoch 7 Results)
-Below are the recorded metrics from the training session:
+### Training & Validation Accuracy Curves
+![Accuracy Curve](https://huggingface.co/Arko007/adaptive-cloud-plant-model/resolve/main/accuracy_curve.png)
 
-| Metric | Score / Value | Description |
-| :--- | :--- | :--- |
-| **Training Loss** | `0.1039` | Cross-Entropy loss at epoch 7 |
-| **Training Accuracy** | `96.47%` | Accuracy over the training split |
-| **Validation Loss** | `0.1028` | Cross-Entropy loss on the held-out split |
-| **Validation Accuracy** | **`96.42%`** | Calibrated validation classification rate |
+---
+
+## 3. Convergence Metrics Summary
+The model was trained for 7 epochs on Kaggle GPU environments using a stratified dataset split (30,103 training images, 5,347 validation images):
+
+| Epoch | Training Loss | Training Accuracy (%) | Validation Loss | Validation Accuracy (%) |
+|:---:|:---:|:---:|:---:|:---:|
+| 1 | 0.6872 | 82.83% | 0.1986 | 93.44% |
+| 2 | 0.2168 | 93.04% | 0.1420 | 95.05% |
+| 3 | 0.1629 | 94.61% | 0.1330 | 95.37% |
+| 4 | 0.1394 | 95.45% | 0.1153 | 95.80% |
+| 5 | 0.1235 | 95.83% | 0.1167 | 96.08% |
+| 6 | 0.1104 | 96.24% | 0.1152 | 95.84% |
+| 7 | **0.1039** | **96.47%** | **0.1028** | **96.42%** |
+
+*Note: Checkpoints for all epochs, including the best-performing `convnext_large_cloud_best.pth`, are stored directly in the Hugging Face model repository.*
+
+---
 
-## Integration & Deployment
-The model checkpoint is designed to be hosted in a FastAPI-based Docker container on Hugging Face Spaces (Port 7860), responding to offloading queries via HTTP REST or binary JSON over MQTT v5 brokers.
+## 4. Collaborative Gating Mechanism
+The Cloud model acts as a secondary diagnostician in the cooperative pipeline. The lightweight edge node decides whether to query this model by checking:
+1. **Evidential Vacuity Threshold ($u > \tau_{vac}$)**: Triggers if the leaf image has out-of-distribution patterns or high epistemic uncertainty.
+2. **Conformal Confidence Threshold ($p_{max} < \tau_{conf}$)**: Triggers if the calibrated categorical prediction confidence is low.