added app.py and others

app.py

@@ -0,0 +1,52 @@
+from model import AQC_NET
+import torch
+import torch.nn as nn
+import torchvision.transforms as T
+from PIL import Image
+import numpy as np
+import gradio as gr
+model = AQC_NET(pretrain=True,num_label=5)
+def predict(image_name):
+    model.eval()
+    
+
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    model.to(device)
+    inputs = preprocess(image_name)
+    inputs = inputs.to(device)
+    with torch.no_grad():
+        outputs = model(inputs.unsqueeze(0))
+        values, indices = torch.topk(outputs, k=5) 
+        print(values,indices)
+    return {i.item(): v.item() for i, v in zip(indices[0], values.detach()[0])}
+def preprocess(image_name):
+    transforms = T.Compose([
+        T.Resize((256,256)),
+        T.ToTensor(),
+        T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
+    ])
+    image = transforms(image_name)
+    return image
+
+def run_gradio():
+    
+    title = "AQC_NET PH"
+    description = "trial AQC_NET"
+    examples = ["test_image.jpg","test_img.jpg"]
+    inputs = [
+        gr.inputs.Image(type="pil", label="Input Image")
+    ]
+
+
+    gr.Interface(
+        predict,
+        inputs,
+        outputs = 'label',
+        title=title,
+        description=description,
+        examples=examples,
+        theme="huggingface",
+    ).launch(debug=True, enable_queue=True)
+
+#print(predict("test_image.jpg"))
+run_gradio()

model.py

@@ -0,0 +1,35 @@
+import torch
+import torch.nn as nn
+from torch import Tensor as tensor
+from torch.nn import functional as F
+import torchvision.models as models
+
+class SCA_Block(nn.Module):
+    def __init__(self, in_channel, downsample_channel):
+        super().__init__()
+        self.conv_A = nn.Conv2d(in_channel, downsample_channel, (1,1))
+        self.conv_B = nn.Conv2d(in_channel, downsample_channel, (1,1))
+        self.conv_E = nn.Conv2d(in_channel, downsample_channel, (1,1))
+        self.linear = nn.Linear(downsample_channel,in_channel)
+    def forward(self, feature_in):
+        b_size,c,w,h = feature_in.shape
+        A = self.conv_A(feature_in)
+        B = self.conv_B(feature_in)
+        E = self.conv_E(feature_in)
+        c1 = A.shape[1]
+        Z = F.softmax(torch.dot(torch.reshape(A,(b_size,c1,-1)),
+                                torch.reshape(B,(b_size,-1,c1))), axis = 1 )
+        D = torch.reshape( Z * torch.reshape(E,(b_size,c1,-1)) , (b_size,c1,w,h))
+        out = feature_in * F.sigmoid(F.adaptive_avg_pool2d(D))
+        return out
+
+
+class AQC_NET(nn.Module):
+    def __init__(self, pretrain = True, num_label = 5):
+        super().__init__()
+        self.resnet18 = models.resnet18(pretrained = pretrain)
+        self.resnet18.layer3[0].add_module('sca_1', SCA_Block(256,16))
+        self.resnet18.layer3[1].add_module('sca_2', SCA_Block(256,16))
+        self.resnet18.fc = nn.Linear(512,num_label)
+    def forward(self,x):
+        return F.softmax(self.resnet18(x))

requirements.txt

@@ -0,0 +1,5 @@
+torch
+torchvision
+pillow
+numpy
+gradio