Update: add sample images selection

app.py

@@ -20,21 +20,26 @@ def process_image(image, yolo_versions=["yolov5"]):
             result_images.append((Image.fromarray(image), f"{yolo_version} not yet implemented."))
     return result_images
 
-
 interface = gr.Interface(
     fn=process_image,
     inputs=[
-        gr.Image(type="pil", label="Upload an Image"),
+        gr.Image(type="pil", label="Upload an Image", optional=True),
         gr.CheckboxGroup(
             choices=["yolov5", "yolov8s"],
             value=["yolov5"],  # Set default selection to YOLOv5
             label="Select Model(s)",
-        )
+        ),
     ],
     outputs=gr.Gallery(label="Results", elem_id="gallery", rows=2, height=500),
     title="Explainable AI for YOLO Models",
-    description="Upload an image to visualize YOLO object detection with Grad-CAM."
+    description="Upload an image or select a sample to visualize YOLO object detection with Grad-CAM.",
+    examples=[
+        ["/data/xai/sample1.jpeg"],  
+        ["/data/xai/sample2.jpeg"],  
+    ],
+    live=True  
 )
 
+
 if __name__ == "__main__":
     interface.launch()

test.py

@@ -1,58 +0,0 @@
-import torch
-from transformers import BertTokenizer, BertForMaskedLM
-import matplotlib.pyplot as plt
-from sklearn.manifold import TSNE
-import numpy as np
-from mpl_toolkits.mplot3d import Axes3D
-
-# Load a pre-trained model and tokenizer
-model_name = 'bert-base-uncased'
-tokenizer = BertTokenizer.from_pretrained(model_name)
-model = BertForMaskedLM.from_pretrained(model_name)
-
-# Example input text
-text = "The quick brown fox jumps over the lazy dog"
-
-# Tokenize the input text
-inputs = tokenizer(text, return_tensors="pt")
-input_ids = inputs['input_ids']
-
-# Get attention weights by running the model
-with torch.no_grad():
-    outputs = model(input_ids, output_attentions=True)
-
-# Extract the attention weights (size: [num_layers, num_heads, seq_len, seq_len])
-attention_weights = outputs.attentions
-
-# Select a specific layer and attention head
-layer_idx = 0  # First layer
-head_idx = 0   # First attention head
-
-# Get the attention matrix for this layer and head
-attention_matrix = attention_weights[layer_idx][0][head_idx].cpu().numpy()
-
-# Use t-SNE to reduce the dimensionality of the attention matrix (embedding space)
-# Attention matrix shape: [seq_len, seq_len], so we reduce each row (which corresponds to a token's attention distribution)
-tsne = TSNE(n_components=3, random_state=42, perplexity=5)  # Set a lower perplexity value
-reduced_attention = tsne.fit_transform(attention_matrix)
-
-# Plotting the reduced attention embeddings in 3D
-fig = plt.figure(figsize=(12, 10))
-ax = fig.add_subplot(111, projection='3d')
-
-# Plot the reduced attention in 3D
-ax.scatter(reduced_attention[:, 0], reduced_attention[:, 1], reduced_attention[:, 2])
-
-# Annotate the tokens in the scatter plot
-tokens = tokenizer.convert_ids_to_tokens(input_ids[0])
-for i, token in enumerate(tokens):
-    ax.text(reduced_attention[i, 0], reduced_attention[i, 1], reduced_attention[i, 2],
-            token, fontsize=12, ha='center')
-
-# Set plot labels
-ax.set_title(f"3D t-SNE Visualization of Attention - Layer {layer_idx+1}, Head {head_idx+1}")
-ax.set_xlabel("t-SNE Dimension 1")
-ax.set_ylabel("t-SNE Dimension 2")
-ax.set_zlabel("t-SNE Dimension 3")
-
-plt.show()