Adapt bounding box annotation to image ratio

app.py

@@ -4,6 +4,7 @@ import gradio as gr
 import supervision as sv
 import torch
 from PIL import Image
+from supervision import Color
 from ultralytics import YOLO
 
 MARKDOWN = """
@@ -12,7 +13,7 @@ MARKDOWN = """
 ## About the model 👁️
 This is a demo for my YOLOv8 nano trained for orang outan detection.\\
 The model was trained using [this dataset](https://images.cv/dataset/orangutan-image-classification-dataset)
-for orang outan images and [this dataset](https://www.kaggle.com/datasets/slothkong/10-monkey-species/data) as background images. Annotations were obtained using zero shot detection method GroundingDino.\\
+for orang outan images and [this dataset](https://www.kaggle.com/datasets/slothkong/10-monkey-species/data) as background images. Annotations were obtained using zero shot object detection method GroundingDino.\\
 
 The code can be found on my github repository: https://github.com/clementapa/orang-outan-image-video-detection.
 
@@ -30,7 +31,7 @@ DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
 YOLO_MODEL = YOLO("train_7best.pt")
 
-BOX_ANNOTATOR = sv.BoxAnnotator()
+BOX_ANNOTATOR = sv.BoxAnnotator(color=Color.from_hex("#FF00E4"))
 
 
 def annotate(
@@ -39,6 +40,20 @@ def annotate(
     annotator: sv.BoxAnnotator,
     labels: str,
 ) -> Image.Image:
+    thickness = 2
+    text_thickness = 1
+    text_scale = 1.0
+
+    height, width, _ = image_bgr_numpy.shape
+
+    thickness_ratio = ((width + height) / 2) / 400
+    text_scale_ratio = ((width + height) / 2) / 600
+    text_thickness_ratio = ((width + height) / 2) / 400
+
+    annotator.thickness = int(thickness * thickness_ratio)
+    annotator.text_scale = float(text_scale * text_scale_ratio)
+    annotator.text_thickness = int(text_thickness * text_thickness_ratio)
+
     annotated_bgr_image = annotator.annotate(
         scene=image_bgr_numpy, detections=detections, labels=labels
     )
@@ -46,7 +61,7 @@ def annotate(
 
 
 def inference(image_rgb_pil: Image.Image, confidence: float) -> List[Image.Image]:
-    output = YOLO_MODEL(image_rgb_pil, verbose=False)[0]
+    output = YOLO_MODEL(image_rgb_pil, imgsz=640, verbose=False)[0]
     detections = sv.Detections.from_ultralytics(output)
 
     detections = detections[detections.confidence >= confidence]