codewithdark
commited on
Update README.md
Browse files
README.md
CHANGED
@@ -54,25 +54,52 @@ The model was fine-tuned using the following settings:
|
|
54 |
To use the fine-tuned model for inference, simply load the model from Hugging Face's Model Hub and input a chest X-ray image:
|
55 |
|
56 |
```python
|
57 |
-
from transformers import ViTForImageClassification, ViTFeatureExtractor
|
58 |
-
import torch
|
59 |
from PIL import Image
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
|
61 |
-
#
|
62 |
-
|
63 |
-
|
64 |
|
65 |
-
#
|
66 |
-
|
|
|
67 |
|
68 |
-
#
|
69 |
-
|
70 |
-
|
71 |
-
logits = outputs.logits
|
72 |
-
predictions = torch.sigmoid(logits).squeeze()
|
73 |
|
74 |
-
|
75 |
-
|
|
|
|
|
|
|
76 |
```
|
77 |
|
78 |
### Fine-Tuning
|
|
|
54 |
To use the fine-tuned model for inference, simply load the model from Hugging Face's Model Hub and input a chest X-ray image:
|
55 |
|
56 |
```python
|
|
|
|
|
57 |
from PIL import Image
|
58 |
+
import torch
|
59 |
+
from transformers import AutoImageProcessor, AutoModelForImageClassification
|
60 |
+
|
61 |
+
# Load model and processor
|
62 |
+
processor = AutoImageProcessor.from_pretrained("codewithdark/vit-chest-xray")
|
63 |
+
model = AutoModelForImageClassification.from_pretrained("codewithdark/vit-chest-xray")
|
64 |
+
|
65 |
+
# Define label columns (class names)
|
66 |
+
label_columns = ['Cardiomegaly', 'Edema', 'Consolidation', 'Pneumonia', 'No Finding']
|
67 |
+
|
68 |
+
# Step 1: Load and preprocess the image
|
69 |
+
image_path = "/content/images.jpeg" # Replace with your image path
|
70 |
+
|
71 |
+
# Open the image
|
72 |
+
image = Image.open(image_path)
|
73 |
+
|
74 |
+
# Ensure the image is in RGB mode (required by most image classification models)
|
75 |
+
if image.mode != 'RGB':
|
76 |
+
image = image.convert('RGB')
|
77 |
+
print("Image converted to RGB.")
|
78 |
+
|
79 |
+
# Step 2: Preprocess the image using the processor
|
80 |
+
inputs = processor(images=image, return_tensors="pt")
|
81 |
+
|
82 |
+
# Step 3: Make a prediction (using the model)
|
83 |
+
with torch.no_grad(): # Disable gradient computation during inference
|
84 |
+
outputs = model(**inputs)
|
85 |
|
86 |
+
# Step 4: Extract logits and get the predicted class index
|
87 |
+
logits = outputs.logits # Raw logits from the model
|
88 |
+
predicted_class_idx = torch.argmax(logits, dim=-1).item() # Get the class index
|
89 |
|
90 |
+
# Step 5: Map the predicted index to a class label
|
91 |
+
# You can also use `model.config.id2label`, but we'll use `label_columns` for this task
|
92 |
+
predicted_class_label = label_columns[predicted_class_idx]
|
93 |
|
94 |
+
# Output the results
|
95 |
+
print(f"Predicted Class Index: {predicted_class_idx}")
|
96 |
+
print(f"Predicted Class Label: {predicted_class_label}")
|
|
|
|
|
97 |
|
98 |
+
'''
|
99 |
+
Output :
|
100 |
+
Predicted Class Index: 4
|
101 |
+
Predicted Class Label: No Finding
|
102 |
+
'''
|
103 |
```
|
104 |
|
105 |
### Fine-Tuning
|