Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch import nn
|
3 |
+
from transformers import SegformerImageProcessor, SegformerForSemanticSegmentation
|
4 |
+
from PIL import Image
|
5 |
+
import matplotlib.pyplot as plt
|
6 |
+
import requests
|
7 |
+
import gradio as gr
|
8 |
+
import numpy as np
|
9 |
+
|
10 |
+
# convenience expression for automatically determining device
|
11 |
+
device = (
|
12 |
+
"cuda"
|
13 |
+
# Device for NVIDIA or AMD GPUs
|
14 |
+
if torch.cuda.is_available()
|
15 |
+
else "mps"
|
16 |
+
# Device for Apple Silicon (Metal Performance Shaders)
|
17 |
+
if torch.backends.mps.is_available()
|
18 |
+
else "cpu"
|
19 |
+
)
|
20 |
+
|
21 |
+
# Load models
|
22 |
+
image_processor = SegformerImageProcessor.from_pretrained("jonathandinu/face-parsing")
|
23 |
+
model = SegformerForSemanticSegmentation.from_pretrained("jonathandinu/face-parsing")
|
24 |
+
model.to(device)
|
25 |
+
|
26 |
+
# Inference function
|
27 |
+
def infer(image: Image.Image) -> np.ndarray:
|
28 |
+
# Preprocess image
|
29 |
+
inputs = image_processor(images=image, return_tensors="pt").to(device)
|
30 |
+
outputs = model(**inputs)
|
31 |
+
logits = outputs.logits # shape (batch_size, num_labels, ~height/4, ~width/4)
|
32 |
+
|
33 |
+
# Resize output to match input image dimensions
|
34 |
+
upsampled_logits = nn.functional.interpolate(logits,
|
35 |
+
size=image.size[::-1], # H x W
|
36 |
+
mode='bilinear',
|
37 |
+
align_corners=False)
|
38 |
+
|
39 |
+
# Get label masks
|
40 |
+
labels = upsampled_logits.argmax(dim=1)[0]
|
41 |
+
|
42 |
+
# Move to CPU to visualize in matplotlib
|
43 |
+
labels_viz = labels.cpu().numpy()
|
44 |
+
return labels_viz
|
45 |
+
|
46 |
+
# Create Gradio interface
|
47 |
+
iface = gr.Interface(
|
48 |
+
fn=infer, # the function to be used for inference
|
49 |
+
inputs=gr.inputs.Image(type="pil"), # input type (image)
|
50 |
+
outputs=gr.outputs.Image(type="numpy"), # output type (image as numpy array)
|
51 |
+
live=True, # run inference live as the image is uploaded
|
52 |
+
title="Face Parsing with Segformer", # interface title
|
53 |
+
description="Upload an image to perform face parsing using the Segformer model for semantic segmentation." # description
|
54 |
+
)
|
55 |
+
|
56 |
+
# Launch the interface
|
57 |
+
iface.launch()
|