Update app.py

app.py

@@ -8,6 +8,7 @@ from PIL import Image
 import itertools
 import numpy as np
 import os
+import logging
 
 
 args = {
@@ -15,23 +16,26 @@ args = {
     "device": torch.device('cuda' if torch.cuda.is_available() else 'cpu'),
     "dxlabels": ["akiec", "bcc", "bkl", "df", "mel", "nv","vasc"]
 }
+logging.warning(args)
 
 # No specific normalization was performed during training
 def normtransform(x):
     return x
 
 # Load model
+logging.warning("Loading model...")
 model = torchvision.models.resnet34()
 model.fc = torch.nn.Linear(model.fc.in_features, len(args.get("dxlabels")))
 model.load_state_dict(torch.load(args.get("model_path")))
 model.eval()
 model.to(device=args.get("device"))
 torch.set_grad_enabled(False)
-
+logging.warning("Model loaded.")
 
 def predict(image: str)->dict:
-    global model, args, normtransform
-    
+    global model, args
+
+    logging.warning(f"Starting predict('{image}') ...")
     prediction_tensor = torch.zeros([1, len(args.get("dxlabels"))]).to(device=args.get("device"))
 
     # Test-time augmentations
@@ -61,6 +65,7 @@ def predict(image: str)->dict:
     
     prediction_tensor /= len(aug_combos)
     predictions = F.softmax(prediction_tensor, dim=1)[0].cpu().numpy()
+    logging.warning(f"Returning {predictions=}")
     return {args.get("dxlabels")[enu]: p for enu, p in enumerate(predictions)}
 
 description = '''Research artifact for multi-class predictions of common 
@@ -71,6 +76,7 @@ For education and research use only.
 **DO NOT use this to obtain medical advice!**
 If you have a skin change in question, seek contact to your physician.'''
 
+logging.warning("Starting Gradio interface...")
 gr.Interface(
     predict,
     inputs=gr.Image(label="Upload a dermatoscopic image", type="filepath"),