Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from PIL import Image
|
3 |
+
from torchvision import transforms
|
4 |
+
import gradio as gr
|
5 |
+
|
6 |
+
torch.hub.download_url_to_file("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
|
7 |
+
|
8 |
+
model = torch.hub.load('pytorch/vision:v0.9.0', 'resnext50_32x4d', pretrained=True)
|
9 |
+
# or
|
10 |
+
# model = torch.hub.load('pytorch/vision:v0.9.0', 'resnext101_32x8d', pretrained=True)
|
11 |
+
model.eval()
|
12 |
+
|
13 |
+
def inference(input_image):
|
14 |
+
preprocess = transforms.Compose([
|
15 |
+
transforms.Resize(256),
|
16 |
+
transforms.CenterCrop(224),
|
17 |
+
transforms.ToTensor(),
|
18 |
+
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
19 |
+
])
|
20 |
+
input_tensor = preprocess(input_image)
|
21 |
+
input_batch = input_tensor.unsqueeze(0) # create a mini-batch as expected by the model
|
22 |
+
|
23 |
+
# move the input and model to GPU for speed if available
|
24 |
+
if torch.cuda.is_available():
|
25 |
+
input_batch = input_batch.to('cuda')
|
26 |
+
model.to('cuda')
|
27 |
+
|
28 |
+
with torch.no_grad():
|
29 |
+
output = model(input_batch)
|
30 |
+
# The output has unnormalized scores. To get probabilities, you can run a softmax on it.
|
31 |
+
probabilities = torch.nn.functional.softmax(output[0], dim=0)
|
32 |
+
# Download ImageNet labels
|
33 |
+
!wget https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt
|
34 |
+
# Read the categories
|
35 |
+
with open("imagenet_classes.txt", "r") as f:
|
36 |
+
categories = [s.strip() for s in f.readlines()]
|
37 |
+
|
38 |
+
# Show top categories per image
|
39 |
+
top5_prob, top5_catid = torch.topk(probabilities, 5)
|
40 |
+
result = {}
|
41 |
+
for i in range(top5_prob.size(0)):
|
42 |
+
result[categories[top5_catid[i]]] = top5_prob[i].item()
|
43 |
+
return result
|
44 |
+
|
45 |
+
inputs = gr.inputs.Image(type='pil')
|
46 |
+
outputs = gr.outputs.Label(type="confidences",num_top_classes=5)
|
47 |
+
|
48 |
+
title = "RESNEXT"
|
49 |
+
description = "Gradio demo for RESNEXT, Next generation ResNets, more efficient and accurate. To use it, simply upload your image, or click one of the examples to load them. Read more at the links below."
|
50 |
+
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/1611.05431'>Aggregated Residual Transformations for Deep Neural Networks</a> | <a href='https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py'>Github Repo</a></p>"
|
51 |
+
|
52 |
+
examples = [
|
53 |
+
['dog.jpg']
|
54 |
+
]
|
55 |
+
gr.Interface(inference, inputs, outputs, title=title, description=description, article=article, examples=examples, analytics_enabled=False).launch()
|