spuuntries
commited on
Commit
·
e8590af
1
Parent(s):
1be0680
feat: add app script
Browse files
app.py
ADDED
|
@@ -0,0 +1,104 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import torch
|
| 3 |
+
import torch.nn as nn
|
| 4 |
+
from torchvision import transforms
|
| 5 |
+
from PIL import Image
|
| 6 |
+
from transformers import ViTForImageClassification, ViTConfig
|
| 7 |
+
import random
|
| 8 |
+
import numpy as np
|
| 9 |
+
import transformers
|
| 10 |
+
from skimage.metrics import structural_similarity as ssim
|
| 11 |
+
import requests
|
| 12 |
+
import os
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
def set_seed(seed):
|
| 16 |
+
random.seed(seed)
|
| 17 |
+
np.random.seed(seed)
|
| 18 |
+
torch.manual_seed(seed)
|
| 19 |
+
torch.cuda.manual_seed_all(seed)
|
| 20 |
+
transformers.set_seed(seed)
|
| 21 |
+
torch.backends.cudnn.deterministic = True
|
| 22 |
+
torch.backends.cudnn.benchmark = False
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
set_seed(42)
|
| 26 |
+
|
| 27 |
+
device = "cpu"
|
| 28 |
+
config = ViTConfig.from_pretrained("google/vit-base-patch16-224")
|
| 29 |
+
config.num_labels = 2 # Binary classification
|
| 30 |
+
|
| 31 |
+
# Download the model file
|
| 32 |
+
model_url = "https://huggingface.co/spuun/yummy-paws/resolve/main/best_model.pth"
|
| 33 |
+
model_path = "best_model.pth"
|
| 34 |
+
|
| 35 |
+
if not os.path.exists(model_path):
|
| 36 |
+
response = requests.get(model_url)
|
| 37 |
+
with open(model_path, "wb") as f:
|
| 38 |
+
f.write(response.content)
|
| 39 |
+
|
| 40 |
+
# Load the trained model
|
| 41 |
+
model = ViTForImageClassification.from_pretrained(
|
| 42 |
+
model_path, config=config, ignore_mismatched_sizes=True
|
| 43 |
+
)
|
| 44 |
+
model.classifier = nn.Linear(model.config.hidden_size, 2)
|
| 45 |
+
model.to(device)
|
| 46 |
+
|
| 47 |
+
# Download the reference image
|
| 48 |
+
reference_image_url = (
|
| 49 |
+
"https://huggingface.co/spuun/yummy-paws/resolve/main/images%20(15).jpeg"
|
| 50 |
+
)
|
| 51 |
+
reference_image_path = "reference_image.jpeg"
|
| 52 |
+
|
| 53 |
+
if not os.path.exists(reference_image_path):
|
| 54 |
+
response = requests.get(reference_image_url)
|
| 55 |
+
with open(reference_image_path, "wb") as f:
|
| 56 |
+
f.write(response.content)
|
| 57 |
+
|
| 58 |
+
# Load the reference image for SSIM comparison
|
| 59 |
+
reference_image = Image.open(reference_image_path)
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
def calculate_ssim(img1, img2):
|
| 63 |
+
img1_array = np.array(img1)
|
| 64 |
+
img2_array = np.array(img2)
|
| 65 |
+
ssim_value = ssim(img1_array, img2_array, channel_axis=2)
|
| 66 |
+
return ssim_value
|
| 67 |
+
|
| 68 |
+
|
| 69 |
+
def predict_and_compare(image):
|
| 70 |
+
image = image.resize(reference_image.size)
|
| 71 |
+
ssim_value = calculate_ssim(image, reference_image)
|
| 72 |
+
|
| 73 |
+
transform = transforms.Compose(
|
| 74 |
+
[
|
| 75 |
+
transforms.Resize((224, 224)),
|
| 76 |
+
transforms.ToTensor(),
|
| 77 |
+
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
|
| 78 |
+
]
|
| 79 |
+
)
|
| 80 |
+
|
| 81 |
+
image_tensor = transform(image).unsqueeze(0).to(device)
|
| 82 |
+
|
| 83 |
+
model.eval()
|
| 84 |
+
with torch.no_grad():
|
| 85 |
+
output = model(image_tensor).logits
|
| 86 |
+
probabilities = torch.softmax(output, dim=1)[0]
|
| 87 |
+
predicted_class_index = torch.argmax(probabilities).item()
|
| 88 |
+
|
| 89 |
+
class_names = ["False", "True"] # Assuming 0 index is False, 1 is True
|
| 90 |
+
predicted_class = class_names[predicted_class_index]
|
| 91 |
+
probability = probabilities[predicted_class_index].item()
|
| 92 |
+
|
| 93 |
+
return f"Predicted: {predicted_class}\nProbability: {probability:.4f}\nSSIM with reference: {ssim_value:.4f}"
|
| 94 |
+
|
| 95 |
+
|
| 96 |
+
iface = gr.Interface(
|
| 97 |
+
fn=predict_and_compare,
|
| 98 |
+
inputs=gr.Image(type="pil"),
|
| 99 |
+
outputs="text",
|
| 100 |
+
title="Image Classification and Comparison",
|
| 101 |
+
description="Upload an image to classify it and compare with a reference image.",
|
| 102 |
+
)
|
| 103 |
+
|
| 104 |
+
iface.launch()
|