Add application file

README.md

@@ -1,10 +1,10 @@
 ---
 title: Deepdespeckling
-emoji: 🦀
-colorFrom: green
-colorTo: gray
+emoji: 🌍
+colorFrom: yellow
+colorTo: red
 sdk: streamlit
-sdk_version: 1.31.1
+sdk_version: 1.31.0
 app_file: app.py
 pinned: false
 license: mit

app.py

@@ -0,0 +1,87 @@
+import streamlit as st
+from PIL import Image
+from io import BytesIO, TextIOWrapper
+import numpy as np
+
+from deepdespeckling.despeckling import get_denoiser, get_model_weights_path
+from deepdespeckling.utils.load_cosar import cos2mat
+from deepdespeckling.utils.constants import PATCH_SIZE, STRIDE_SIZE
+
+st.set_page_config(layout="wide", page_title="Deepdespeckling")
+
+st.write("## Despeckle your SAR images")
+st.write(
+    "Try to upload a light image "
+)
+st.sidebar.write("## Upload and download :gear:")
+
+
+def convert_image(img):
+    buf = BytesIO()
+    img.save(buf, format="PNG")
+    byte_im = buf.getvalue()
+    return byte_im
+
+
+def preprocess_for_png(image, threshold):
+    noisy_image = np.clip(image, 0, threshold)
+    noisy_image = noisy_image / threshold * 255
+    noisy_image = Image.fromarray(image.astype('float64')).convert('L')
+
+    return noisy_image
+
+
+def preprocess_noisy_image(upload_path, denoiser):
+    image = cos2mat(upload_path)
+    noisy_image = np.array(image).reshape(
+        1, np.size(image, 0), np.size(image, 1), 2)
+    noisy_image, _, _ = denoiser.preprocess_noisy_image(noisy_image)
+    threshold = np.mean(noisy_image) + 3 * np.std(noisy_image)
+
+    noisy_image = preprocess_for_png(noisy_image, threshold=threshold)
+
+    return image, noisy_image, threshold
+
+
+def fix_image(upload_path):
+    model_name = "spotlight"
+    denoiser = get_denoiser(model_name=model_name)
+
+    image, noisy_image, threshold = preprocess_noisy_image(
+        upload_path=upload_path, denoiser=denoiser)
+
+    col1.write("Original Image :camera:")
+    col1.image(noisy_image)
+
+    model_weights_path = get_model_weights_path(model_name=model_name)
+    despeckled_image = denoiser.denoise_image(
+        image, model_weights_path, PATCH_SIZE, STRIDE_SIZE)["denoised"]["full"]
+
+    despeckled_image = preprocess_for_png(
+        despeckled_image, threshold=threshold)
+
+    col2.write("Despeckled Image")
+    col2.image(despeckled_image)
+    st.sidebar.markdown("\n")
+    st.sidebar.download_button("Download despeckled image", convert_image(
+        despeckled_image), "despeckled.png", "image/png")
+
+
+def init_image():
+    col1.write("Noisy Image ")
+    col1.image("img/noisy.png")
+
+    col1.write("Despeckled Image")
+    col1.image("img/denoised.png")
+
+
+col1, col2 = st.columns(2)
+my_upload = st.sidebar.file_uploader("Upload an image", type=["cos"])
+
+if my_upload is not None:
+    b = my_upload.getvalue()
+    with open("test.cos", "wb") as f:
+        f.write(b)
+    fix_image("test.cos")
+else:
+    init_image()

packages.txt

@@ -0,0 +1 @@
+libgdal-dev

requirements.txt

@@ -0,0 +1,4 @@
+deepdespeckling
+pillow
+numpy
+GDAL==3.6.2