dcrey7 commited on
Commit
9caacdc
·
verified ·
1 Parent(s): 61fd66a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +58 -0
README.md CHANGED
@@ -11,4 +11,62 @@ license: apache-2.0
11
  short_description: image segmentation
12
  ---
13
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14
  Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
11
  short_description: image segmentation
12
  ---
13
 
14
+
15
+ # Wetlands Segmentation App
16
+
17
+ This Hugging Face Space provides an interactive web interface for segmenting wetland areas in satellite imagery using a DeepLabv3+ model.
18
+
19
+ ## Features
20
+
21
+ - Upload satellite imagery in common formats (JPG, PNG) or GeoTIFF format
22
+ - Optionally upload ground truth masks for evaluation
23
+ - Visualize wetland segmentation predictions
24
+ - Calculate metrics (IoU, Precision, Recall, F1) when ground truth is provided
25
+ - View wetland coverage percentage statistics
26
+
27
+ ## Usage Instructions
28
+
29
+ 1. **Upload Input Image**:
30
+ - Use the "Upload Image" tab for common image formats (JPG, PNG, etc.)
31
+ - Use the "Upload TIFF" tab for GeoTIFF files with multiple bands
32
+
33
+ 2. **Upload Ground Truth (Optional)**:
34
+ - If you have a ground truth mask, upload it to see evaluation metrics
35
+ - The ground truth should be a binary mask where white (255) represents wetlands
36
+
37
+ 3. **Analyze**:
38
+ - Click the "Analyze Image" button to process the image
39
+ - View the segmentation results and statistics
40
+
41
+ ## Model Information
42
+
43
+ This app uses a model from the [dcrey7/wetlands_segmentation_deeplabsv3plus](https://huggingface.co/dcrey7/wetlands_segmentation_deeplabsv3plus) repository.
44
+
45
+ **Model Architecture:**
46
+ - DeepLabv3+ with ResNet-50 backbone
47
+ - Input: RGB satellite imagery (optimal size: 128×128 pixels)
48
+ - Output: Binary segmentation mask (Wetland vs Background)
49
+
50
+ The model was trained on a dataset of satellite imagery containing wetland regions, focusing on environmental monitoring and conservation planning applications.
51
+
52
+ ## Example Output
53
+
54
+ When you upload an image, the app will display:
55
+ - The original input image
56
+ - The predicted wetland segmentation mask
57
+ - Ground truth mask (if provided)
58
+ - Statistics including wetland coverage percentage
59
+ - Evaluation metrics (if ground truth is provided)
60
+
61
+ ## Limitations
62
+
63
+ - The model works best on imagery similar to its training data
64
+ - Performance may vary depending on image quality and characteristics
65
+ - The model is designed for 128×128 pixel inputs (images will be resized)
66
+ - While the model can process images with any number of bands, it was trained on RGB data
67
+
68
+ ## License
69
+
70
+ This application and the underlying model are available under the Apache 2.0 license.
71
+
72
  Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference