Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,73 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
#app.py:
|
3 |
+
# from huggingface_hub import from_pretrained_fastai
|
4 |
+
import gradio as gr
|
5 |
+
from fastcore.xtras import Path
|
6 |
+
from fastai.callback.hook import summary
|
7 |
+
from fastai.callback.progress import ProgressCallback
|
8 |
+
from fastai.callback.schedule import lr_find, fit_flat_cos
|
9 |
+
from fastai.data.block import DataBlock
|
10 |
+
from fastai.data.external import untar_data, URLs
|
11 |
+
from fastai.data.transforms import get_image_files, FuncSplitter, Normalize
|
12 |
+
from fastai.layers import Mish
|
13 |
+
from fastai.losses import BaseLoss
|
14 |
+
from fastai.optimizer import ranger
|
15 |
+
from fastai.torch_core import tensor
|
16 |
+
from fastai.vision.augment import aug_transforms
|
17 |
+
from fastai.vision.core import PILImage, PILMask
|
18 |
+
from fastai.vision.data import ImageBlock, MaskBlock, imagenet_stats
|
19 |
+
from fastai.vision.learner import unet_learner
|
20 |
+
from PIL import Image
|
21 |
+
import numpy as np
|
22 |
+
from torch import nn
|
23 |
+
from torchvision.models.resnet import resnet34
|
24 |
+
import torch
|
25 |
+
import torch.nn.functional as F
|
26 |
+
|
27 |
+
|
28 |
+
# # repo_id = "YOUR_USERNAME/YOUR_LEARNER_NAME"
|
29 |
+
repo_id = "islasher/segm-grapes"
|
30 |
+
|
31 |
+
|
32 |
+
# # Definimos una función que se encarga de llevar a cabo las predicciones
|
33 |
+
|
34 |
+
|
35 |
+
# # Cargar el modelo y el tokenizador
|
36 |
+
learn = load_learner(repo_id)
|
37 |
+
#learner = from_pretrained_fastai(repo_id)
|
38 |
+
|
39 |
+
import torchvision.transforms as transforms
|
40 |
+
def transform_image(image):
|
41 |
+
my_transforms = transforms.Compose([transforms.ToTensor(),
|
42 |
+
transforms.Normalize(
|
43 |
+
[0.485, 0.456, 0.406],
|
44 |
+
[0.229, 0.224, 0.225])])
|
45 |
+
image_aux = image
|
46 |
+
return my_transforms(image_aux).unsqueeze(0).to(device)
|
47 |
+
|
48 |
+
|
49 |
+
|
50 |
+
|
51 |
+
|
52 |
+
|
53 |
+
# Definimos una función que se encarga de llevar a cabo las predicciones
|
54 |
+
def predict(img):
|
55 |
+
image = transforms.Resize((480,640))(img)
|
56 |
+
tensor = transform_image(image=image)
|
57 |
+
with torch.no_grad():
|
58 |
+
outputs = model(tensor)
|
59 |
+
|
60 |
+
outputs = torch.argmax(outputs,1)
|
61 |
+
|
62 |
+
mask = np.array(outputs.cpu())
|
63 |
+
mask[mask==1]=150
|
64 |
+
mask[mask==3]=76 #pole # y no 74
|
65 |
+
# mask[mask==5]=74 #pole
|
66 |
+
mask[mask==2]=29 #wood # y no 25
|
67 |
+
# mask[mask==6]=25 #wood
|
68 |
+
mask[mask==4]=255 #grape
|
69 |
+
mask=np.reshape(mask,(480,640)) #en modo matriz
|
70 |
+
return Image.fromarray(mask.astype('uint8'))
|
71 |
+
|
72 |
+
# Creamos la interfaz y la lanzamos.
|
73 |
+
gr.Interface(fn=predict, inputs=gr.inputs.Image(shape=(128, 128)), outputs=gr.outputs.Image(shape=(480,640)),examples=['color_154.jpg','color_155.jpg']).launch(share=False)
|