Spaces:
Sleeping
Sleeping
azhongai666666
commited on
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch import nn
|
3 |
+
import torchvision
|
4 |
+
import numpy as np
|
5 |
+
import matplotlib.pyplot as plt
|
6 |
+
from PIL import Image
|
7 |
+
import pandas as pd
|
8 |
+
import segmentation_models_pytorch as smp
|
9 |
+
import gradio as gr
|
10 |
+
|
11 |
+
num_classes = 2
|
12 |
+
model_unet_path = "unet_model.pth"
|
13 |
+
model_fpn_path = "fpn_model.pth"
|
14 |
+
model_deeplab_path = "deeplabv3_model.pth"
|
15 |
+
image_path = "leaf11.jpg"
|
16 |
+
|
17 |
+
# Get cpu or gpu device for training.
|
18 |
+
device = "cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
|
19 |
+
print(f"Using {device} device")
|
20 |
+
|
21 |
+
model_unet = smp.Unet(
|
22 |
+
encoder_name="resnet18", # choose encoder, e.g. mobilenet_v2 or efficientnet-b7
|
23 |
+
encoder_weights=None, # use `imagenet` pre-trained weights for encoder initialization
|
24 |
+
in_channels=3, # model input channels (1 for gray-scale images, 3 for RGB, etc.)
|
25 |
+
classes=num_classes, # model output channels (number of classes in your dataset)
|
26 |
+
)
|
27 |
+
|
28 |
+
model_fpn = smp.FPN(
|
29 |
+
encoder_name="resnet18", # choose encoder, e.g. mobilenet_v2 or efficientnet-b7
|
30 |
+
encoder_weights=None, # use `imagenet` pre-trained weights for encoder initialization
|
31 |
+
in_channels=3, # model input channels (1 for gray-scale images, 3 for RGB, etc.)
|
32 |
+
classes=num_classes, # model output channels (number of classes in your dataset)
|
33 |
+
)
|
34 |
+
|
35 |
+
model_deeplab = smp.DeepLabV3(
|
36 |
+
encoder_name="resnet34", # choose encoder, e.g. mobilenet_v2 or efficientnet-b7
|
37 |
+
encoder_weights=None, # use `imagenet` pre-trained weights for encoder initialization
|
38 |
+
in_channels=3, # model input channels (1 for gray-scale images, 3 for RGB, etc.)
|
39 |
+
classes=num_classes, # model output channels (number of classes in your dataset)
|
40 |
+
)
|
41 |
+
|
42 |
+
def pred_one_image(inp,option):
|
43 |
+
one_image = np.array(inp.resize((256, 256)).convert("RGB"))
|
44 |
+
# convert to other format HWC -> CHW
|
45 |
+
one_image = np.moveaxis(one_image, -1, 0)
|
46 |
+
# mask = np.expand_dims(mask, 0)
|
47 |
+
one_image = torch.tensor(one_image).float()
|
48 |
+
one_image = one_image.unsqueeze(0)
|
49 |
+
one_image = one_image.to(device)
|
50 |
+
if option == "unet":
|
51 |
+
model_load = model_unet
|
52 |
+
elif option == "fpn":
|
53 |
+
model_load = model_fpn
|
54 |
+
elif option == "deeplab":
|
55 |
+
model_load = model_deeplab
|
56 |
+
model_load.eval()
|
57 |
+
with torch.no_grad():
|
58 |
+
output = model_load(one_image)
|
59 |
+
# print(output.shape)
|
60 |
+
predictions = torch.argmax(output, dim=1) # 获取预测的类别标签图像
|
61 |
+
pred_array = (predictions[0].cpu().numpy()/2*255).astype(np.uint8)
|
62 |
+
# print(pred_array.shape)
|
63 |
+
pred_img = Image.fromarray(pred_array)
|
64 |
+
# pred_img.save("pred.png")
|
65 |
+
# print(predictions.shape)
|
66 |
+
return pred_img
|
67 |
+
|
68 |
+
|
69 |
+
|
70 |
+
model_unet.load_state_dict(torch.load(model_unet_path,map_location=torch.device('cpu')))
|
71 |
+
model_fpn.load_state_dict(torch.load(model_fpn_path,map_location=torch.device('cpu')))
|
72 |
+
model_deeplab.load_state_dict(torch.load(model_deeplab_path,map_location=torch.device('cpu')))
|
73 |
+
|
74 |
+
dropdown = gr.Dropdown(["unet", "fpn","deeplab"])
|
75 |
+
interface = gr.Interface(fn=pred_one_image,
|
76 |
+
inputs=[gr.Image(type="pil"),dropdown],
|
77 |
+
outputs=gr.Image(type="pil"),
|
78 |
+
examples=[["leaf11.jpg",'unet']],)
|
79 |
+
interface.launch(debug=False)
|