view file

.all-contributorsrc

@@ -0,0 +1,35 @@
+{
+  "files": [
+    "README.md"
+  ],
+  "imageSize": 100,
+  "commit": false,
+  "commitType": "docs",
+  "commitConvention": "angular",
+  "contributors": [
+    {
+      "login": "abhijitjadhav1998",
+      "name": "Abhijit Jadhav",
+      "avatar_url": "https://avatars.githubusercontent.com/u/38549908?v=4",
+      "profile": "https://www.linkedin.com/in/abhijitjadhav1998/",
+      "contributions": [
+        "projectManagement"
+      ]
+    },
+    {
+      "login": "vthonte",
+      "name": "Vishwesh Thonte",
+      "avatar_url": "https://avatars.githubusercontent.com/u/43621438?v=4",
+      "profile": "http://vthonte.vercel.app/",
+      "contributions": [
+        "maintenance"
+      ]
+    }
+  ],
+  "contributorsPerLine": 7,
+  "skipCi": true,
+  "repoType": "github",
+  "repoHost": "https://github.com",
+  "projectName": "Deepfake_detection_using_deep_learning",
+  "projectOwner": "abhijitjadhav1998"
+}

Django Application/README.md

@@ -36,7 +36,7 @@ docker run -p 80:80 --volumes-from deepfakeapplication -v static_volume:/home/ap
 
 ### Step 5: Star⭐ this repo 😉 on <a href="https://github.com/abhijitjadhav1998/Deepfake_detection_using_deep_learning" >  <img src="https://img.shields.io/badge/GitHub-100000?style=for-the-badge&logo=github&logoColor=white" /> </a> and   Star⭐ this image on <a href="https://hub.docker.com/r/abhijitjadhav1998/deefake-detection-20framemodel">  <img src="https://img.shields.io/badge/Docker-2CA5E0?style=for-the-badge&logo=docker&logoColor=white" /> </a>
 
-## We deserve a Coffee ☕ <a href="https://www.buymeacoffee.com/abhijitjadhav" target="_blank"><img src="https://www.buymeacoffee.com/assets/img/custom_images/orange_img.png" alt="Buy Me A Coffee" style="height: 35px !important;width: 174px !important;box-shadow: 0px 3px 2px 0px rgba(190, 190, 190, 0.5) !important;-webkit-box-shadow: 0px 3px 2px 0px rgba(190, 190, 190, 0.5) !important;" ></a> <a href="https://github.com/abhijitjadhav1998/Deepfake_detection_using_deep_learning" target="_blank"><img src="https://img.shields.io/badge/UPI-abhijit.jadhav1%40ybl-brightgreen" alt="Medium" style="height: 35px !important;width: 250px !important;"/></a>&nbsp;
+## We deserve a Coffee ☕ <a href="https://www.buymeacoffee.com/abhijitjadhav" target="_blank"><img src="https://www.buymeacoffee.com/assets/img/custom_images/orange_img.png" alt="Buy Me A Coffee" style="height: 35px !important;width: 174px !important;box-shadow: 0px 3px 2px 0px rgba(190, 190, 190, 0.5) !important;-webkit-box-shadow: 0px 3px 2px 0px rgba(190, 190, 190, 0.5) !important;" ></a>
 
 
 Please note that currently we have only pushed the image of 20 Frames model, If you can to create your own image of other frames model follow the steps given in the [blog](https://abhijithjadhav.medium.com/dockerise-deepfake-detection-django-application-using-nvidia-cuda-40cdda3b6d38).

Django Application/ml_app/views.py

@@ -25,12 +25,17 @@
 
 index_template_name = 'index.html'
 predict_template_name = 'predict.html'
+about_template_name = "about.html"
 
 im_size = 112
 mean=[0.485, 0.456, 0.406]
 std=[0.229, 0.224, 0.225]
 sm = nn.Softmax()
 inv_normalize =  transforms.Normalize(mean=-1*np.divide(mean,std),std=np.divide([1,1,1],std))
+if torch.cuda.is_available():
+    device = 'gpu'
+else:
+    device = 'cpu'
 
 train_transforms = transforms.Compose([
                                         transforms.ToPILImage(),
@@ -124,12 +129,12 @@ def im_plot(tensor):
     image = cv2.merge((r,g,b))
     image = image*[0.22803, 0.22145, 0.216989] +  [0.43216, 0.394666, 0.37645]
     image = image*255.0
-    plt.imshow(image.astype(int))
+    plt.imshow(image.astype('uint8'))
     plt.show()
 
 
 def predict(model,img,path = './', video_file_name=""):
-  fmap,logits = model(img.to('cuda'))
+  fmap,logits = model(img.to(device))
   img = im_convert(img[:,-1,:,:,:], video_file_name)
   params = list(model.parameters())
   weight_softmax = model.linear1.weight.detach().cpu().numpy()
@@ -140,7 +145,7 @@ def predict(model,img,path = './', video_file_name=""):
   return [int(prediction.item()),confidence]
 
 def plot_heat_map(i, model, img, path = './', video_file_name=''):
-  fmap,logits = model(img.to('cuda'))
+  fmap,logits = model(img.to(device))
   params = list(model.parameters())
   weight_softmax = model.linear1.weight.detach().cpu().numpy()
   logits = sm(logits)
@@ -173,25 +178,30 @@ def get_accurate_model(sequence_length):
     sequence_model = []
     final_model = ""
     list_models = glob.glob(os.path.join(settings.PROJECT_DIR, "models", "*.pt"))
-    for i in list_models:
-        model_name.append(i.split("\\")[-1])
-    for i in model_name:
+
+    for model_path in list_models:
+        model_name.append(os.path.basename(model_path))
+
+    for model_filename in model_name:
         try:
-            seq = i.split("_")[3]
-            if (int(seq) == sequence_length):
-                sequence_model.append(i)
-        except:
-            pass
+            seq = model_filename.split("_")[3]
+            if int(seq) == sequence_length:
+                sequence_model.append(model_filename)
+        except IndexError:
+            pass  # Handle cases where the filename format doesn't match expected
 
     if len(sequence_model) > 1:
         accuracy = []
-        for i in sequence_model:
-            acc = i.split("_")[1]
-            accuracy.append(acc)
+        for filename in sequence_model:
+            acc = filename.split("_")[1]
+            accuracy.append(acc)  # Convert accuracy to float for proper comparison
         max_index = accuracy.index(max(accuracy))
-        final_model = sequence_model[max_index]
+        final_model = os.path.join(settings.PROJECT_DIR, "models", sequence_model[max_index])
+    elif len(sequence_model) == 1:
+        final_model = os.path.join(settings.PROJECT_DIR, "models", sequence_model[0])
     else:
-        final_model = sequence_model[0]
+        print("No model found for the specified sequence length.")  # Handle no models found case
+
     return final_model
 
 ALLOWED_VIDEO_EXTENSIONS = set(['mp4','gif','webm','avi','3gp','wmv','flv','mkv'])
@@ -248,114 +258,130 @@ def index(request):
 
 def predict_page(request):
     if request.method == "GET":
+        # Redirect to 'home' if 'file_name' is not in session
         if 'file_name' not in request.session:
             return redirect("ml_app:home")
         if 'file_name' in request.session:
             video_file = request.session['file_name']
         if 'sequence_length' in request.session:
             sequence_length = request.session['sequence_length']
         path_to_videos = [video_file]
-        video_file_name = video_file.split('\\')[-1]
-        if settings.DEBUG == False:
-            production_video_name = video_file_name.split('/')[3:]
-            production_video_name = '/'.join([str(elem) for elem in production_video_name])
-            print("Production file name",production_video_name)
-        video_file_name_only = video_file_name.split('.')[0]
-        video_dataset = validation_dataset(path_to_videos, sequence_length=sequence_length,transform= train_transforms)
-        model = Model(2).cuda()
-        model_name = os.path.join(settings.PROJECT_DIR,'models', get_accurate_model(sequence_length))
-        models_location = os.path.join(settings.PROJECT_DIR,'models')
+        video_file_name = os.path.basename(video_file)
+        video_file_name_only = os.path.splitext(video_file_name)[0]
+        # Production environment adjustments
+        if not settings.DEBUG:
+            production_video_name = os.path.join('/home/app/staticfiles/', video_file_name.split('/')[3])
+            print("Production file name", production_video_name)
+        else:
+            production_video_name = video_file_name
+
+        # Load validation dataset
+        video_dataset = validation_dataset(path_to_videos, sequence_length=sequence_length, transform=train_transforms)
+
+        # Load model
+        if(device == "gpu"):
+            model = Model(2).cuda()  # Adjust the model instantiation according to your model structure
+        else:
+            model = Model(2).cpu()  # Adjust the model instantiation according to your model structure
+        model_name = os.path.join(settings.PROJECT_DIR, 'models', get_accurate_model(sequence_length))
         path_to_model = os.path.join(settings.PROJECT_DIR, model_name)
-        model.load_state_dict(torch.load(path_to_model))
+        model.load_state_dict(torch.load(path_to_model, map_location=torch.device('cpu')))
         model.eval()
         start_time = time.time()
-        # Start: Displaying preprocessing images
+        # Display preprocessing images
         print("<=== | Started Videos Splitting | ===>")
         preprocessed_images = []
         faces_cropped_images = []
         cap = cv2.VideoCapture(video_file)
-
         frames = []
-        while(cap.isOpened()):
+        while cap.isOpened():
             ret, frame = cap.read()
-            if ret==True:
+            if ret:
                 frames.append(frame)
-                if cv2.waitKey(1) & 0xFF == ord('q'):
-                    break
             else:
                 break
         cap.release()
 
-        for i in range(1, sequence_length+1):
-            frame = frames[i]
-            image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
-            img = pImage.fromarray(image, 'RGB')
-            image_name = video_file_name_only+"_preprocessed_"+str(i)+'.png'
-            if settings.DEBUG:
-                image_path = os.path.join(settings.PROJECT_DIR, 'uploaded_images', image_name)
-            else:
-                print("image_name",image_name)
-                image_path = "/home/app/staticfiles" + image_name
-            img.save(image_path)
-            preprocessed_images.append(image_name)
-        print("<=== | Videos Splitting Done | ===>")
-        print("--- %s seconds ---" % (time.time() - start_time))
-        # End: Displaying preprocessing images
-
-
-        # Start: Displaying Faces Cropped Images
-        print("<=== | Started Face Cropping Each Frame | ===>")
+        print(f"Number of frames: {len(frames)}")
+        # Process each frame for preprocessing and face cropping
         padding = 40
         faces_found = 0
-        for i in range(1, sequence_length+1):
+        for i in range(sequence_length):
+            if i >= len(frames):
+                break
             frame = frames[i]
-            #fig, ax = plt.subplots(1,1, figsize=(5, 5))
-            face_locations = face_recognition.face_locations(frame)
+
+            # Convert BGR to RGB
+            rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+
+            # Save preprocessed image
+            image_name = f"{video_file_name_only}_preprocessed_{i+1}.png"
+            image_path = os.path.join(settings.PROJECT_DIR, 'uploaded_images', image_name)
+            img_rgb = pImage.fromarray(rgb_frame, 'RGB')
+            img_rgb.save(image_path)
+            preprocessed_images.append(image_name)
+
+            # Face detection and cropping
+            face_locations = face_recognition.face_locations(rgb_frame)
             if len(face_locations) == 0:
                 continue
+
             top, right, bottom, left = face_locations[0]
-            frame_face = frame[top-padding:bottom+padding, left-padding:right+padding]
-            image = cv2.cvtColor(frame_face, cv2.COLOR_BGR2RGB)
+            frame_face = frame[top - padding:bottom + padding, left - padding:right + padding]
 
-            img = pImage.fromarray(image, 'RGB')
-            image_name = video_file_name_only+"_cropped_faces_"+str(i)+'.png'
-            if settings.DEBUG:
-                image_path = os.path.join(settings.PROJECT_DIR, 'uploaded_images', video_file_name_only+"_cropped_faces_"+str(i)+'.png')
-            else:
-                image_path = "/home/app/staticfiles" + image_name
-            img.save(image_path)
-            faces_found = faces_found + 1
+            # Convert cropped face image to RGB and save
+            rgb_face = cv2.cvtColor(frame_face, cv2.COLOR_BGR2RGB)
+            img_face_rgb = pImage.fromarray(rgb_face, 'RGB')
+            image_name = f"{video_file_name_only}_cropped_faces_{i+1}.png"
+            image_path = os.path.join(settings.PROJECT_DIR, 'uploaded_images', image_name)
+            img_face_rgb.save(image_path)
+            faces_found += 1
             faces_cropped_images.append(image_name)
-        print("<=== | Face Cropping Each Frame Done | ===>")
+
+        print("<=== | Videos Splitting and Face Cropping Done | ===>")
         print("--- %s seconds ---" % (time.time() - start_time))
 
-        # No face is detected
+        # No face detected
         if faces_found == 0:
-            return render(request, predict_template_name, {"no_faces": True})
+            return render(request, 'predict_template_name.html', {"no_faces": True})
 
-        # End: Displaying Faces Cropped Images
+        # Perform prediction
         try:
             heatmap_images = []
-            for i in range(0, len(path_to_videos)):
-                output = ""
-                print("<=== | Started Predicition | ===>")
+            output = ""
+            confidence = 0.0
+
+            for i in range(len(path_to_videos)):
+                print("<=== | Started Prediction | ===>")
                 prediction = predict(model, video_dataset[i], './', video_file_name_only)
                 confidence = round(prediction[1], 1)
-                print("<=== |  Predicition Done | ===>")
-                # print("<=== | Heat map creation started | ===>")
-                # for j in range(0, sequence_length):
-                #     heatmap_images.append(plot_heat_map(j, model, video_dataset[i], './', video_file_name_only))
-                if prediction[0] == 1:
-                    output = "REAL"
-                else:
-                    output = "FAKE"
-                print("Prediction : " , prediction[0],"==",output ,"Confidence : " , confidence)
+                output = "REAL" if prediction[0] == 1 else "FAKE"
+                print("Prediction:", prediction[0], "==", output, "Confidence:", confidence)
+                print("<=== | Prediction Done | ===>")
                 print("--- %s seconds ---" % (time.time() - start_time))
+
+                # Uncomment if you want to create heat map images
+                # for j in range(sequence_length):
+                #     heatmap_images.append(plot_heat_map(j, model, video_dataset[i], './', video_file_name_only))
+
+            # Render results
+            context = {
+                'preprocessed_images': preprocessed_images,
+                'faces_cropped_images': faces_cropped_images,
+                'heatmap_images': heatmap_images,
+                'original_video': production_video_name,
+                'models_location': os.path.join(settings.PROJECT_DIR, 'models'),
+                'output': output,
+                'confidence': confidence
+            }
+
             if settings.DEBUG:
-                return render(request, predict_template_name, {'preprocessed_images': preprocessed_images, 'heatmap_images': heatmap_images, "faces_cropped_images": faces_cropped_images, "original_video": video_file_name, "models_location": models_location, "output": output, "confidence": confidence})
+                return render(request, predict_template_name, context)
             else:
-                return render(request, predict_template_name, {'preprocessed_images': preprocessed_images, 'heatmap_images': heatmap_images, "faces_cropped_images": faces_cropped_images, "original_video": production_video_name, "models_location": models_location, "output": output, "confidence": confidence})
-        except:
+                return render(request, predict_template_name, context)
+
+        except Exception as e:
+            print(f"Exception occurred during prediction: {e}")
             return render(request, 'cuda_full.html')
 def about(request):
     return render(request, about_template_name)