Add merged model + processor

handler.py

@@ -5,10 +5,34 @@ import io
 from typing import Any, Dict
 
 import torch
+import torchvision.transforms as T
 from PIL import Image
 from transformers import AutoImageProcessor, Dinov2ForImageClassification
 
-from train_model import get_inference_transform
+
+def get_inference_transform(processor: AutoImageProcessor, size: int):
+    """Get the raw validation transform for direct inference on PIL images."""
+    normalize   = T.Normalize(mean=processor.image_mean, std=processor.image_std)
+
+    to_rgb = T.Lambda(lambda img: img.convert('RGB'))
+ 
+    def pad_to_square(img):
+        w, h = img.size
+        max_size = max(w, h)
+        pad_w = (max_size - w) // 2
+        pad_h = (max_size - h) // 2
+        padding = (pad_w, pad_h, max_size - w - pad_w, max_size - h - pad_h)
+        return T.Pad(padding, fill=0)(img)
+
+    aug     = T.Compose([
+        to_rgb,
+        pad_to_square,
+        T.Resize(size),
+        T.ToTensor(), 
+        normalize
+    ])
+
+    return aug
 
 
 class EndpointHandler:

requirements.txt

@@ -1,5 +1,2 @@
 torchvision>=0.19
-Pillow>=10
-datasets>=2.19
-peft>=0.10
-safetensors>=0.4
+Pillow>=10