normalize logits

run.py

@@ -3,7 +3,7 @@ import numpy as np
 from time import sleep
 import torch
 from transformers import SegformerImageProcessor, SegformerForSemanticSegmentation
-from torchvision import transforms
+# from torchvision import transforms
 
 weights2load = 'segformer_ep15_loss0.00.pth'
 id2label = {0: 'seal', 255: 'bck'}
@@ -32,12 +32,13 @@ def flip_periodically(im, interval_ms=0):
         The flipped image.
     """
     
-    transforms.ToTensor()(im)
+    # transforms.ToTensor()(im)
     pixel_values = image_processor(im, return_tensors="pt").pixel_values.to(device)
     outputs = model(pixel_values=pixel_values)
     logits = outputs.logits.cpu().detach().numpy()
     sleep(interval_ms / 1000)  # Convert milliseconds to seconds
-    return logits[0, 0]  #np.flipud(im)
+    imout = (logits[0, 0] - logits[0, 0].min()) / (logits[0, 0].max() - logits[0, 0].min())
+    return imout  #np.flipud(im)
 
 with gr.Blocks() as demo:
     inp = gr.Image(sources=["webcam"], streaming=True)