Upload folder using huggingface_hub

README.md

@@ -1,3 +1,66 @@
----
-license: mit
----
+---
+license: mit
+tags: 
+- vision
+- image-classification
+---
+
+# Kindwise Crossroad Model
+
+## The model use it
+
+Here is how to use this model to classify an image into one of the basic classes:
+
+```python
+from huggingface_hub import hf_hub_download
+import cv2
+import numpy as np
+import PIL.Image
+import torch
+import torchvision
+
+DEVICE_NAME = 'cuda:0'
+MODEL_PATH = hf_hub_download('kindwise/crossroad.tiny', 'model.traced.pt')
+CLASSES_PATH = hf_hub_download('kindwise/crossroad.tiny', 'classes.txt')
+IMAGE_PATH = '/tmp/photo.jpg' 
+
+with open(CLASSES_PATH) as f:
+    CLASSES = [line.strip() for line in f]
+MODEL = torch.jit.load(MODEL_PATH).eval().to(DEVICE_NAME)
+
+
+def resize_crop(image_data: np.ndarray, target_size: int = 480) -> np.ndarray | None:
+    height, width, _ = image_data.shape
+    # Determine the size of the square crop
+    crop_size = min(height, width)
+    # Calculate coordinates for center crop
+    start_x = (width - crop_size) // 2
+    start_y = (height - crop_size) // 2
+    # Perform center crop
+    cropped_img = image_data[
+        start_y : start_y + crop_size,
+        start_x : start_x + crop_size
+    ]
+    # Resize cropped image to target size
+    return  cv2.resize(
+        cropped_img,
+        (target_size, target_size),
+        interpolation=cv2.INTER_AREA,
+    )
+
+with torch.no_grad():
+    image_array = np.array(PIL.Image.open(IMAGE_PATH))
+    image_array_resized = resize_crop(image_array)
+    image_tensor = torchvision.transforms.functional.to_tensor(image_array_resized).to(DEVICE_NAME)
+    prediction = MODEL(image_tensor.unsqueeze(0)).squeeze(0).cpu().numpy()
+    for i in (-prediction).argsort():
+        print(f'{classes[i]:>10}: {100 * prediction[i]:.1f}%')
+```
+
+Output:
+```
+     plant: 96.7%
+     human: 0.1%
+    insect: 0.1%
+  mushroom: 0.0%
+```