Update app.py

app.py

@@ -50,8 +50,8 @@ def clip_optimized_latent(text, seed, iterations=25, lr=0.1):
     params = [torch.nn.Parameter(latent_vector[i], requires_grad=True) for i in range(len(latent_vector))]
     optimizer = Adam(params, lr=lr)
 
-    #with torch.no_grad():
-    #    text_features = clip_model.encode_text(text_input)
+    with torch.no_grad():
+        text_features = clip_model.encode_text(text_input)
         
     #pbar = tqdm(range(iterations), dynamic_ncols=True)
     
@@ -67,16 +67,16 @@ def clip_optimized_latent(text, seed, iterations=25, lr=0.1):
         #image = clip_preprocess(Image.fromarray((image_np * 255).astype(np.uint8))).unsqueeze(0).to(device)
         
         # Extract features from the image
-        #image_features = clip_model.encode_image(image)
+        image_features = clip_model.encode_image(image)
 
         # Calculate the loss and backpropagate
-        loss = 1 - clip_model(image, text_input)[0] / 100
-        #loss = -torch.cosine_similarity(text_features, image_features).mean()
+        loss = -torch.cosine_similarity(text_features, image_features).mean()
         loss.backward()
         optimizer.step()
         
         #pbar.set_description(f"Loss: {loss.item()}")  # Update the progress bar to show the current loss
-        w = [param.detach().cpu().numpy() for param in params]
+    print(f"Loss: {loss.item()}")   
+    w = [param.detach().cpu().numpy() for param in params]
     
     return w