Update README.md

README.md

@@ -1,344 +1,66 @@
 ---
 license: mit
 ---
-# Evaluation Report for yolov9e_bb_detect_model
+# Model Card: yolov9e_bb_detect_model
 
-**Tasks:** Single-class Object Detection, Feature extraction
+## Model Details
+- **Model Name:** `yolov9e_bb_detect_model`
+- **Model Type:** Single-Class Object Detection and Feature Extractor
+- **Description:** This model is designed to detect the presence of bark beetles in images. It identifies and places a bounding box around the target but does not classify different species of bark beetles. It operates under the single class label: **'bark_beetle'**.
 
-## Evaluation Notes
-
-This model does not identify individual species but detects a single category of object.
-
-The evaluation was performed on a single-class basis using the text prompt: **'bark_beetle'**.
-
-### Mantel Correlation Explanation
-
-The Mantel R statistic is calculated by comparing the distances between clustering centroids of different species to their phylogenetic distances. This helps determine if the model's learned feature representations correlate with the evolutionary relationships between species.
-
-## Object Classification Performance
-
-**mAP@[.5:.95]:** 0.945
-
-### mAP per IoU Threshold
-
-| IoU Threshold   |      mAP |
-|:----------------|---------:|
-| [email protected]        | 0.985164 |
-| [email protected]        | 0.984198 |
-| [email protected]        | 0.983082 |
-| [email protected]        | 0.982256 |
-| [email protected]        | 0.981038 |
-| [email protected]        | 0.977971 |
-| [email protected]        | 0.973752 |
-| [email protected]        | 0.965713 |
-| [email protected]        | 0.951162 |
-| [email protected]        | 0.670233 |
-
-### Average Precision per Class (at last IoU threshold)
-
-| Class       |       AP |
-|:------------|---------:|
-| bark_beetle | 0.670233 |
-
-### Classification Metrics per IoU Threshold
-
-#### IoU Threshold: iou_0.50
-
-- **Accuracy:** 0.988
-- **Balanced Accuracy:** 0.988
-- **Macro Precision:** 1.000
-- **Macro Recall:** 0.988
-- **Macro F1 Score:** 0.994
-- **Cohen's Kappa:** 0.000
-- **Matthews Corrcoef:** 0.000
-
-##### Confusion Matrix
-
-```
-Predicted Label      0
-True Label            
-0                16289
-```
-
-##### Classification Report
-
-```
-              precision   recall  f1-score  support
-0                   1.0  0.98841  0.994171  16480.0
-micro avg           1.0  0.98841  0.994171  16480.0
-macro avg           1.0  0.98841  0.994171  16480.0
-weighted avg        1.0  0.98841  0.994171  16480.0
-```
-
-#### IoU Threshold: iou_0.55
-
-- **Accuracy:** 0.987
-- **Balanced Accuracy:** 0.987
-- **Macro Precision:** 1.000
-- **Macro Recall:** 0.987
-- **Macro F1 Score:** 0.994
-- **Cohen's Kappa:** 0.000
-- **Matthews Corrcoef:** 0.000
-
-##### Confusion Matrix
-
-```
-Predicted Label      0
-True Label            
-0                16269
-```
-
-##### Classification Report
-
-```
-              precision    recall  f1-score  support
-0                   1.0  0.987197  0.993557  16480.0
-micro avg           1.0  0.987197  0.993557  16480.0
-macro avg           1.0  0.987197  0.993557  16480.0
-weighted avg        1.0  0.987197  0.993557  16480.0
-```
-
-#### IoU Threshold: iou_0.60
-
-- **Accuracy:** 0.986
-- **Balanced Accuracy:** 0.986
-- **Macro Precision:** 1.000
-- **Macro Recall:** 0.986
-- **Macro F1 Score:** 0.993
-- **Cohen's Kappa:** 0.000
-- **Matthews Corrcoef:** 0.000
-
-##### Confusion Matrix
-
-```
-Predicted Label      0
-True Label            
-0                16242
-```
-
-##### Classification Report
-
-```
-              precision    recall  f1-score  support
-0                   1.0  0.985558  0.992727  16480.0
-micro avg           1.0  0.985558  0.992727  16480.0
-macro avg           1.0  0.985558  0.992727  16480.0
-weighted avg        1.0  0.985558  0.992727  16480.0
-```
-
-#### IoU Threshold: iou_0.65
-
-- **Accuracy:** 0.984
-- **Balanced Accuracy:** 0.984
-- **Macro Precision:** 1.000
-- **Macro Recall:** 0.984
-- **Macro F1 Score:** 0.992
-- **Cohen's Kappa:** 0.000
-- **Matthews Corrcoef:** 0.000
-
-##### Confusion Matrix
-
-```
-Predicted Label      0
-True Label            
-0                16222
-```
-
-##### Classification Report
-
-```
-              precision    recall  f1-score  support
-0                   1.0  0.984345  0.992111  16480.0
-micro avg           1.0  0.984345  0.992111  16480.0
-macro avg           1.0  0.984345  0.992111  16480.0
-weighted avg        1.0  0.984345  0.992111  16480.0
-```
-
-#### IoU Threshold: iou_0.70
-
-- **Accuracy:** 0.983
-- **Balanced Accuracy:** 0.983
-- **Macro Precision:** 1.000
-- **Macro Recall:** 0.983
-- **Macro F1 Score:** 0.991
-- **Cohen's Kappa:** 0.000
-- **Matthews Corrcoef:** 0.000
-
-##### Confusion Matrix
-
-```
-Predicted Label      0
-True Label            
-0                16201
-```
-
-##### Classification Report
-
-```
-              precision   recall  f1-score  support
-0                   1.0  0.98307  0.991463  16480.0
-micro avg           1.0  0.98307  0.991463  16480.0
-macro avg           1.0  0.98307  0.991463  16480.0
-weighted avg        1.0  0.98307  0.991463  16480.0
-```
-
-#### IoU Threshold: iou_0.75
-
-- **Accuracy:** 0.980
-- **Balanced Accuracy:** 0.980
-- **Macro Precision:** 1.000
-- **Macro Recall:** 0.980
-- **Macro F1 Score:** 0.990
-- **Cohen's Kappa:** 0.000
-- **Matthews Corrcoef:** 0.000
-
-##### Confusion Matrix
-
-```
-Predicted Label      0
-True Label            
-0                16154
-```
-
-##### Classification Report
-
-```
-              precision    recall  f1-score  support
-0                   1.0  0.980218   0.99001  16480.0
-micro avg           1.0  0.980218   0.99001  16480.0
-macro avg           1.0  0.980218   0.99001  16480.0
-weighted avg        1.0  0.980218   0.99001  16480.0
-```
-
-#### IoU Threshold: iou_0.80
-
-- **Accuracy:** 0.977
-- **Balanced Accuracy:** 0.977
-- **Macro Precision:** 1.000
-- **Macro Recall:** 0.977
-- **Macro F1 Score:** 0.988
-- **Cohen's Kappa:** 0.000
-- **Matthews Corrcoef:** 0.000
-
-##### Confusion Matrix
-
-```
-Predicted Label      0
-True Label            
-0                16093
-```
-
-##### Classification Report
-
-```
-              precision    recall  f1-score  support
-0                   1.0  0.976517  0.988119  16480.0
-micro avg           1.0  0.976517  0.988119  16480.0
-macro avg           1.0  0.976517  0.988119  16480.0
-weighted avg        1.0  0.976517  0.988119  16480.0
-```
-
-#### IoU Threshold: iou_0.85
-
-- **Accuracy:** 0.969
-- **Balanced Accuracy:** 0.969
-- **Macro Precision:** 1.000
-- **Macro Recall:** 0.969
-- **Macro F1 Score:** 0.985
-- **Cohen's Kappa:** 0.000
-- **Matthews Corrcoef:** 0.000
-
-##### Confusion Matrix
-
-```
-Predicted Label      0
-True Label            
-0                15977
-```
-
-##### Classification Report
-
-```
-              precision    recall  f1-score  support
-0                   1.0  0.969478  0.984503  16480.0
-micro avg           1.0  0.969478  0.984503  16480.0
-macro avg           1.0  0.969478  0.984503  16480.0
-weighted avg        1.0  0.969478  0.984503  16480.0
-```
-
-#### IoU Threshold: iou_0.90
-
-- **Accuracy:** 0.957
-- **Balanced Accuracy:** 0.957
-- **Macro Precision:** 1.000
-- **Macro Recall:** 0.957
-- **Macro F1 Score:** 0.978
-- **Cohen's Kappa:** 0.000
-- **Matthews Corrcoef:** 0.000
-
-##### Confusion Matrix
+---
 
-```
-Predicted Label      0
-True Label            
-0                15765
-```
+## Evaluation Datasets
 
-##### Classification Report
+To understand the model's capabilities, its performance was tested on two different types of datasets:
 
-```
-              precision    recall  f1-score  support
-0                   1.0  0.956614  0.977826  16480.0
-micro avg           1.0  0.956614  0.977826  16480.0
-macro avg           1.0  0.956614  0.977826  16480.0
-weighted avg        1.0  0.956614  0.977826  16480.0
-```
+-   **In-Distribution (ID):** This dataset contains images that are **similar to the data the model was trained on**. Performance on this dataset shows how well the model performs on familiar types of images.
+-   **Out-of-Distribution (OOD):** This dataset contains images that are **intentionally different species from the training data**.
 
-#### IoU Threshold: iou_0.95
+---
 
-- **Accuracy:** 0.743
-- **Balanced Accuracy:** 0.743
-- **Macro Precision:** 1.000
-- **Macro Recall:** 0.743
-- **Macro F1 Score:** 0.853
-- **Cohen's Kappa:** 0.000
-- **Matthews Corrcoef:** 0.000
+## Performance
 
-##### Confusion Matrix
+### Object Detection
+The model's ability to correctly identify and locate bark beetles is measured by its **mean Average Precision (mAP)**. This metric evaluates both the accuracy of the bounding box placement and the classification confidence. The score is averaged over multiple Intersection over Union (IoU) thresholds, from 50% overlap (`0.50`) to 95% overlap (`0.95`), providing a comprehensive view of prediction accuracy. A higher mAP score indicates better performance.
 
-```
-Predicted Label      0
-True Label            
-0                12252
-```
+| Dataset | mAP (0.50 : 0.95) | Notes |
+| :--- | :--- | :--- |
+| **In-Distribution (ID)** | 🟩 0.9609 | Shows outstanding detection accuracy on images similar to its training data. |
+| **Out-of-Distribution (OOD)**| 🟦 0.9447 | Maintains high accuracy on novel species, demonstrating strong generalization. |
 
-##### Classification Report
+<br>
 
-```
-              precision    recall  f1-score  support
-0                   1.0  0.743447  0.852847  16480.0
-micro avg           1.0  0.743447  0.852847  16480.0
-macro avg           1.0  0.743447  0.852847  16480.0
-weighted avg        1.0  0.743447  0.852847  16480.0
-```
+### Feature Extraction (Embedding Performance)
+The model can also convert images into numerical representations (embeddings). The quality of these embeddings is evaluated by how well they group similar species together in a feature space.
 
-## Embedding Quality
+#### Internal Cluster Validation
+These metrics measure the quality of the clusters formed by the embeddings without referring to ground-truth labels. They assess how dense and well-separated the clusters are.
 
-### Internal Cluster Validation
+| Metric | ID Score | OOD Score | Interpretation |
+| :--- | :--- | :--- | :--- |
+| **Silhouette Score** | 0.6419 | 0.3866 | Measures how similar an object is to its own cluster compared to others. **Higher is better (closer to 1)**. The ID embeddings form better-defined clusters. |
+| **Davies-Bouldin Index**| 0.3203 | 0.3103 | Measures the average similarity between each cluster and its most similar one. **Lower is better (closer to 0)**. Both datasets show a low degree of cluster overlap. |
+| **Calinski-Harabasz Index**| 3127.21 | 721.657 | Measures the ratio of between-cluster dispersion to within-cluster dispersion. **Higher is better**. The ID embeddings form significantly denser and more separated clusters. |
 
-|   Silhouette_Score |   Davies-Bouldin_Index |   Calinski-Harabasz_Index |
-|-------------------:|-----------------------:|--------------------------:|
-|           0.646403 |               0.406169 |                     53572 |
+#### External Cluster Validation
+These metrics evaluate the clustering performance by comparing the results to the true species labels.
 
-### External Cluster Validation
+| Metric | ID Score | OOD Score | Interpretation |
+| :--- | :--- | :--- | :--- |
+| **Adjusted Rand Index (ARI)** | 0.0965 | 0.0047 | Measures the similarity between true and predicted labels, correcting for chance. **Higher is better (closer to 1)**. |
+| **Normalized Mutual Info (NMI)** | 0.3698 | 0.2633 | Measures the agreement between the clustering and the true labels. **Higher is better (closer to 1)**. |
+| **Cluster Purity** | 0.2473 | 0.1246 | Measures the extent to which clusters contain a single class. **Higher is better (closer to 1)**. |
 
-|        ARI |       NMI |   Cluster_Purity |
-|-----------:|----------:|-----------------:|
-| 0.00297115 | 0.0383566 |        0.0771425 |
+**Conclusion:** The external validation scores are low for both datasets, indicating the model's feature representations do **not** effectively separate different species of bark beetles on their own.
 
-### Mantel Correlation
+#### Phylogenetic Correlation (Mantel Test)
+This test determines if the model's learned features correlate with the evolutionary relationships (phylogeny) between different bark beetle species.
 
-|          r |   p_value |   n_items |
-|-----------:|----------:|----------:|
-| -0.0637179 |      0.66 |        32 |
+-   **Mantel R-statistic:** This value ranges from -1 to 1. A positive value means species that are close in the model's feature space are also close evolutionarily. A value near zero indicates no correlation.
+-   **p-value:** Indicates the statistical significance of the result. A p-value below 0.05 typically suggests a significant correlation.
 
+| Dataset | Mantel R-statistic | p-value | Interpretation |
+| :--- | :--- | :--- | :--- |
+| **In-Distribution (ID)** | 0.0621 | 0.5850 | There is **no statistically significant correlation** between the model's feature embeddings and the species' evolutionary history. |
+| **Out-of-Distribution (OOD)**| 0.0388 | 0.6530 | There is **no statistically significant correlation** for the OOD data either. |