Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -5,7 +5,7 @@ from torchvision import transforms
|
|
| 5 |
from torchvision.models import swin_t
|
| 6 |
from PIL import Image
|
| 7 |
|
| 8 |
-
#
|
| 9 |
class MMIM(nn.Module):
|
| 10 |
def __init__(self, num_classes=36):
|
| 11 |
super(MMIM, self).__init__()
|
|
@@ -22,7 +22,7 @@ class MMIM(nn.Module):
|
|
| 22 |
features = self.backbone(x)
|
| 23 |
return self.classifier(features)
|
| 24 |
|
| 25 |
-
#
|
| 26 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 27 |
model = MMIM(num_classes=36)
|
| 28 |
checkpoint = torch.load("MMIM_best.pth", map_location=device)
|
|
@@ -33,7 +33,7 @@ model.load_state_dict(filtered_checkpoint, strict=False)
|
|
| 33 |
model.to(device)
|
| 34 |
model.eval()
|
| 35 |
|
| 36 |
-
#
|
| 37 |
class_names = [
|
| 38 |
"Chinee apple", # class1
|
| 39 |
"Black grass", # class14
|
|
@@ -74,53 +74,53 @@ class_names = [
|
|
| 74 |
"Snake weed",
|
| 75 |
]
|
| 76 |
|
| 77 |
-
#
|
| 78 |
weed_info = {
|
| 79 |
-
"Chinee apple": "
|
| 80 |
-
"Black grass": "
|
| 81 |
-
"Charlock": "
|
| 82 |
-
"Cleavers": "
|
| 83 |
-
"Common Chickweed": "
|
| 84 |
-
"Common Wheat": "
|
| 85 |
-
"Fat Hen": "
|
| 86 |
-
"Lanthana": "
|
| 87 |
-
"Loose Silky bent": "
|
| 88 |
-
"Maize": "
|
| 89 |
-
"Scentless Mayweed": "
|
| 90 |
-
"Shepherds Purse": "
|
| 91 |
-
"Small-Flowered Cranesbill": "
|
| 92 |
-
"Sugar beet": "
|
| 93 |
-
"Carpetweeds": "
|
| 94 |
-
"Crabgrass": "
|
| 95 |
-
"Eclipta": "
|
| 96 |
-
"Goosegrass": "
|
| 97 |
-
"Negative": "
|
| 98 |
-
"Morningglory": "
|
| 99 |
-
"Nutsedge": "
|
| 100 |
-
"Palmer Amarnath": "
|
| 101 |
-
"Prickly Sida": "
|
| 102 |
-
"Purslane": "
|
| 103 |
-
"Ragweed": "
|
| 104 |
-
"Sicklepod": "
|
| 105 |
-
"SpottedSpurge": "
|
| 106 |
-
"SpurredAnoda": "
|
| 107 |
-
"Swinecress": "
|
| 108 |
-
"Parkinsonia": "
|
| 109 |
-
"Waterhemp": "
|
| 110 |
-
"Parthenium": "
|
| 111 |
-
"Prickly acacia": "
|
| 112 |
-
"Rubber vine": "
|
| 113 |
-
"Siam weed": "
|
| 114 |
-
"Snake weed": "
|
| 115 |
}
|
| 116 |
|
| 117 |
-
#
|
| 118 |
transform = transforms.Compose([
|
| 119 |
transforms.Resize((224, 224)),
|
| 120 |
transforms.ToTensor()
|
| 121 |
])
|
| 122 |
|
| 123 |
-
#
|
| 124 |
def predict(img):
|
| 125 |
img = img.convert('RGB')
|
| 126 |
img_tensor = transform(img).unsqueeze(0).to(device)
|
|
@@ -134,35 +134,35 @@ def predict(img):
|
|
| 134 |
confidence = conf.item() * 100
|
| 135 |
|
| 136 |
if predicted_class.lower() == "negative":
|
| 137 |
-
label = f"
|
| 138 |
elif confidence < 60:
|
| 139 |
-
label = f"
|
| 140 |
else:
|
| 141 |
-
label = f"
|
| 142 |
|
| 143 |
-
info = weed_info.get(predicted_class, "
|
| 144 |
-
return f"{label}\n\n
|
| 145 |
|
| 146 |
-
#
|
| 147 |
about_markdown = """
|
| 148 |
-
###
|
| 149 |
This tool predicts weed species from images using a Vision Transformer backbone trained with multi-masked image modeling.
|
| 150 |
|
| 151 |
-
-
|
| 152 |
-
-
|
| 153 |
-
-
|
| 154 |
-
-
|
| 155 |
|
| 156 |
> Tip: Use clear, focused weed images for better results.
|
| 157 |
"""
|
| 158 |
|
| 159 |
-
#
|
| 160 |
interface = gr.Interface(
|
| 161 |
fn=predict,
|
| 162 |
inputs=gr.Image(type="pil", label="Upload an Image"),
|
| 163 |
outputs=gr.Textbox(label="Prediction"),
|
| 164 |
-
title="
|
| 165 |
-
description="A
|
| 166 |
article=about_markdown
|
| 167 |
)
|
| 168 |
|
|
|
|
| 5 |
from torchvision.models import swin_t
|
| 6 |
from PIL import Image
|
| 7 |
|
| 8 |
+
# Model definition
|
| 9 |
class MMIM(nn.Module):
|
| 10 |
def __init__(self, num_classes=36):
|
| 11 |
super(MMIM, self).__init__()
|
|
|
|
| 22 |
features = self.backbone(x)
|
| 23 |
return self.classifier(features)
|
| 24 |
|
| 25 |
+
# Load model
|
| 26 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 27 |
model = MMIM(num_classes=36)
|
| 28 |
checkpoint = torch.load("MMIM_best.pth", map_location=device)
|
|
|
|
| 33 |
model.to(device)
|
| 34 |
model.eval()
|
| 35 |
|
| 36 |
+
# Class names
|
| 37 |
class_names = [
|
| 38 |
"Chinee apple", # class1
|
| 39 |
"Black grass", # class14
|
|
|
|
| 74 |
"Snake weed",
|
| 75 |
]
|
| 76 |
|
| 77 |
+
# Weed info dictionary
|
| 78 |
weed_info = {
|
| 79 |
+
"Chinee apple": " Invasive shrub. Control by uprooting or herbicide treatment.",
|
| 80 |
+
"Black grass": " Infests cereal crops. Remove before seed shedding.",
|
| 81 |
+
"Charlock": " Common weed in oilseed crops. Responds to early herbicide.",
|
| 82 |
+
"Cleavers": " Sticky climbing weed. Control before flowering.",
|
| 83 |
+
"Common Chickweed": " Fast-spreading groundcover weed. Avoid soil disturbance.",
|
| 84 |
+
"Common Wheat": " May appear as weed in rotation crops.",
|
| 85 |
+
"Fat Hen": " Broadleaf weed. Competes heavily with crops.",
|
| 86 |
+
"Lanthana": " Invasive ornamental plant, toxic to livestock.",
|
| 87 |
+
"Loose Silky bent": " Grass weed affecting wheat fields.",
|
| 88 |
+
"Maize": " Sometimes emerges as volunteer weed post-harvest.",
|
| 89 |
+
"Scentless Mayweed": " Strong competitor in cereals. Shallow-rooted.",
|
| 90 |
+
"Shepherds Purse": " Common weed in cool seasons. Heart-shaped pods.",
|
| 91 |
+
"Small-Flowered Cranesbill": " Low-growing, thrives in dry areas.",
|
| 92 |
+
"Sugar beet": " Appears as volunteer in crop fields.",
|
| 93 |
+
"Carpetweeds": " Low mat-forming weed. Easy to remove manually.",
|
| 94 |
+
"Crabgrass": " Summer annual grass. Thrives in disturbed soil.",
|
| 95 |
+
"Eclipta": " Moisture-loving herbaceous weed.",
|
| 96 |
+
"Goosegrass": " Mat-forming weed, tough to hand-pull.",
|
| 97 |
+
"Negative": " No weed confidently detected. Please recheck input.",
|
| 98 |
+
"Morningglory": " Climbing vine, chokes crops quickly.",
|
| 99 |
+
"Nutsedge": " Grass-like weed with tubers. Hard to control.",
|
| 100 |
+
"Palmer Amarnath": " Highly aggressive and herbicide-resistant.",
|
| 101 |
+
"Prickly Sida": " Hairy, thorny stems. Requires early control.",
|
| 102 |
+
"Purslane": " Succulent weed, common in warm climates.",
|
| 103 |
+
"Ragweed": " Allergen-producing weed. Kill before flowering.",
|
| 104 |
+
"Sicklepod": " Toxic to livestock. Control before pod set.",
|
| 105 |
+
"SpottedSpurge": " Low-growing. Releases milky sap.",
|
| 106 |
+
"SpurredAnoda": " Fast-growing summer annual. Common in cotton.",
|
| 107 |
+
"Swinecress": " Strong odor. Grows in compacted soils.",
|
| 108 |
+
"Parkinsonia": " Woody shrub. Mechanical removal advised.",
|
| 109 |
+
"Waterhemp": " Fast-growing amaranth. Glyphosate-resistant strains exist.",
|
| 110 |
+
"Parthenium": " Toxic and invasive. Avoid contact.",
|
| 111 |
+
"Prickly acacia": " Thorny shrub. Displaces native plants.",
|
| 112 |
+
"Rubber vine": " Woody climber. Toxic and invasive.",
|
| 113 |
+
"Siam weed": " Highly invasive in tropical zones.",
|
| 114 |
+
"Snake weed": " Woody perennial, toxic to livestock."
|
| 115 |
}
|
| 116 |
|
| 117 |
+
# Transform
|
| 118 |
transform = transforms.Compose([
|
| 119 |
transforms.Resize((224, 224)),
|
| 120 |
transforms.ToTensor()
|
| 121 |
])
|
| 122 |
|
| 123 |
+
# Prediction function
|
| 124 |
def predict(img):
|
| 125 |
img = img.convert('RGB')
|
| 126 |
img_tensor = transform(img).unsqueeze(0).to(device)
|
|
|
|
| 134 |
confidence = conf.item() * 100
|
| 135 |
|
| 136 |
if predicted_class.lower() == "negative":
|
| 137 |
+
label = f" Predicted as: Negative\nConfidence: {confidence:.2f}%"
|
| 138 |
elif confidence < 60:
|
| 139 |
+
label = f" Low confidence. Possibly Not a Weed\nConfidence: {confidence:.2f}%"
|
| 140 |
else:
|
| 141 |
+
label = f" Predicted class: {predicted_class}\nConfidence: {confidence:.2f}%"
|
| 142 |
|
| 143 |
+
info = weed_info.get(predicted_class, " No additional info available.")
|
| 144 |
+
return f"{label}\n\n Info: {info}"
|
| 145 |
|
| 146 |
+
# App description
|
| 147 |
about_markdown = """
|
| 148 |
+
### Weed Classifier β Swin Transformer + MMIM
|
| 149 |
This tool predicts weed species from images using a Vision Transformer backbone trained with multi-masked image modeling.
|
| 150 |
|
| 151 |
+
- Shows confidence scores
|
| 152 |
+
- Flags uncertain or non-weed predictions
|
| 153 |
+
- Displays weed info after prediction
|
| 154 |
+
- Upload an image
|
| 155 |
|
| 156 |
> Tip: Use clear, focused weed images for better results.
|
| 157 |
"""
|
| 158 |
|
| 159 |
+
# Gradio Interface
|
| 160 |
interface = gr.Interface(
|
| 161 |
fn=predict,
|
| 162 |
inputs=gr.Image(type="pil", label="Upload an Image"),
|
| 163 |
outputs=gr.Textbox(label="Prediction"),
|
| 164 |
+
title=" Weed Image Classifier",
|
| 165 |
+
description="A Self- Spervised Learning model for weed image classification.",
|
| 166 |
article=about_markdown
|
| 167 |
)
|
| 168 |
|