Update README.md

README.md

@@ -11,6 +11,9 @@ base_model:
   This model aims to help count and identify species at a specific methane seep site called Pythia’s Oasis, off the coast of Oregon. The site is at a depth of about 1,000 meters and is unique because of the presence of warm diffuse flow as well as methane hydrate. The site has a high diversity and density of marine life, and is distinct enough from nearby sites that it requires its own model. 
 
 ---
+## Model Details
+This model was trained using YOLOv12 for object detection.
+
 ## Dataset
 This dataset is a series of frames pulled from ROV video during the Pythia’s Oasis research cruise in 2019. Although multiple dives occurred in this location, this dataset only includes imagery from the first dive, J2-1225. The original dive footage was edited down to about an hour and a half, excluding the ascent and descent, as well as removing video with the ROV arm or other machinery. In total, 1,110 images were extracted from the edited footage. Initial annotation predictions were made by Fathomnet, and classes were then cleaned manually. The classes include a variety of anemones, fishes, crabs, corals, sea stars, jellies, and sea cucumbers. Clams and snails, while present, were excluded due to time constraints during annotation.
 | Class Name             | Number of Annotations |
@@ -39,4 +42,7 @@ This dataset is a series of frames pulled from ROV video during the Pythia’s O
 ---
 ## Model Performance
 The model currently confuses many classes for the background and misses many classes. This is because of incomplete annotation of all classes, and likely because clams and snails were excluded. Additionally, classes were very unbalanced. Classes with less annotations tended to have either extreme high or low F1 scores, and some classes were not included in the val set (Black Coral). Continued class cleaning and supplementation with additional imagery should improve this model in later versions. Despite these issues with the current  model, the F1 score is fairly good on average. Rockfish, Actiniaria, and Scotoplanes performed well in terms of having a high F1 score across a large confidence range, and the number of annotations was high for these classes, so this is more likely to reflect accuracy in the model predictions.
-  
+
+---
+## Potential Use Case
+This model could be used to characterize species richness, diversity and density along a video transect. This data could then be compared to other methane seep sites, or compared to previous surveys if the site is surveyed again in the future. An example hypothesis that could be addressed with this model would be: Pythia's Oasis has unique species density and diversity compared to other methane seep sites.