added new samples

app.py

@@ -1,11 +1,63 @@
-import os
 import streamlit as st
 import zipfile
-import torch
 from utils import *
+import numpy as np
 import matplotlib.pyplot as plt
+import matplotlib.animation as animation
+import streamlit.components.v1 as components
 from matplotlib import colors
 
+st.set_page_config(layout="wide")
+
+def create_animation(images, pred_dates):
+    print('Creating composition of images...')
+    fps = 2
+    fig_an, ax_an = plt.subplots()
+    plt.title("")
+    a = images[0]
+    im = ax_an.imshow(a, interpolation='none', aspect='auto', vmin=0, vmax=1)
+
+    title = ax_an.text(0.5, 0.85, "", bbox={'facecolor': 'w', 'alpha': 0.5, 'pad': 5},
+                    transform=ax_an.transAxes, ha="center")
+
+    def animate_func(idx):
+        title.set_text("date: " + pred_dates[idx])
+        im.set_array(images[idx])
+        return [im]
+
+    anima = animation.FuncAnimation(fig_an, animate_func, frames=len(images), interval=1000 / fps, blit=True,
+                                    repeat=False)
+    print('Done!')
+    return anima
+
+
+def load_daily_preds_as_animations(pred_full_paths, pred_dates):
+    daily_preds = []
+    for path in pred_full_paths:
+        img, _ = read(path)
+        img = np.squeeze(img)
+        img = [classes_color_map[p] for p in img]
+        daily_preds.append(img)
+    anima = create_animation(daily_preds, pred_dates)
+    return anima
+
+
+def load_src_images_as_animations(img_paths, pred_dates):
+    imgs = []
+    for path in img_paths:
+        img, _ = read(path)
+
+        # https://custom-scripts.sentinel-hub.com/custom-scripts/sentinel-2/composites/
+        # IREA image:
+        # False colors (8,4,3): 2,blue-B3,green-B4,5,6,7,red-B8,11,12
+        # Simple RGB (4, 3, 2): blue-B2,green-B3,red-B4,5,6,7,8,11,12
+        rgb = img[[2, 1, 0], :, :]
+        rgb = np.moveaxis(rgb, 0, -1)
+        imgs.append(rgb/np.amax(rgb))
+    anima = create_animation(imgs, pred_dates)
+    return anima
+
+
 if not hasattr(st, 'paths'):
     st.paths = None
 if not hasattr(st, 'daily_model'):
@@ -43,7 +95,7 @@ if not hasattr(st, 'daily_model'):
 
 
 st.title('In-season and dynamic crop mapping using 3D convolution neural networks and sentinel-2 time series')
-st.markdown(""" Demo App for the model presented in the paper:
+st.markdown(""" Demo App for the model presented in the [paper](https://www.sciencedirect.com/science/article/pii/S0924271622003203):
 ```
 @article{gallo2022in_season,
   title = {In-season and dynamic crop mapping using 3D convolution neural networks and sentinel-2 time series},
@@ -57,6 +109,7 @@ st.markdown(""" Demo App for the model presented in the paper:
   author = {Ignazio Gallo and Luigi Ranghetti and Nicola Landro and Riccardo {La Grassa} and Mirco Boschetti},
 }
 ```
+***The demo doesn't work properly, we are working to fix the bugs!*
 """)
 
 file_uploaded = st.file_uploader(
@@ -65,6 +118,7 @@ file_uploaded = st.file_uploader(
     accept_multiple_files=False,
 )
 sample_path = None
+tileids = None
 if file_uploaded is not None:
     with zipfile.ZipFile(file_uploaded, "r") as z:
         z.extractall("uploaded_samples")
@@ -72,10 +126,20 @@ if file_uploaded is not None:
 st.markdown('or use a demo sample')
 if st.button('sample 1'):
     sample_path = 'demo_data/lombardia'
+    tileids = ['24']
+if st.button('sample 2'):
+    sample_path = 'demo_data/lombardia'
+    tileids = ['712']
+if st.button('sample 3'):
+    sample_path = 'demo_data/lombardia'
+    tileids = ['814']
+if st.button('sample 4'):
+    sample_path = 'demo_data/lombardia'
+    tileids = ['1509']
 
 paths = None
 if sample_path is not None:
-    st.markdown(f'elaborating {sample_path}...')
+    # st.markdown(f'elaborating {sample_path} ...')
 
     validationdataset = SentinelDailyAnnualDatasetNoLabel(
         sample_path,
@@ -83,11 +147,11 @@ if sample_path is not None:
         opt.classes_path,
         opt.sample_duration,
         opt.win_size,
-        tileids=None)
+        tileids=tileids)
     validationdataloader = torch.utils.data.DataLoader(
         validationdataset, batch_size=opt.batch_size, shuffle=False, num_workers=opt.workers)
 
-    st.markdown(f'predict in progress...')
+    st.markdown('Model prediction in progress ...')
 
     out_dir = os.path.join(opt.result_path, "seg_maps")
     if not os.path.exists(out_dir):
@@ -154,7 +218,7 @@ if sample_path is not None:
 
                     # save GT image in  opt.root_path
                     full_pth_date = os.path.join(
-                        full_pth_patch, dates[ch][batch] + f'-ch{ch}-b{batch}-daily-pred.tif')
+                        full_pth_patch, dates[ch] + '-daily-pred.tif')
                     profile.update({
                         'nodata': None,
                         'dtype': 'uint8',
@@ -164,15 +228,21 @@ if sample_path is not None:
 
     st.markdown('End prediction')
 
-    folder = "demo_data/results/seg_maps/example-lombardia/2"
-    st.paths = os.listdir(folder)
+    # folder_out = "demo_data/results/seg_maps/example-lombardia/2"
+    folder_out = os.path.join("demo_data/results/seg_maps/"+opt.years[0]+"-lombardia/", tileids[0])
+    st.paths = os.listdir(folder_out)
+    st.paths.sort()
 
 if st.paths is not None:
-    folder = "demo_data/results/seg_maps/example-lombardia/2"
-    file_picker = st.selectbox("Select day predict (annual is patch-pred-nn.tif)",
-                               st.paths, index=st.paths.index('patch-pred-nn.tif'))
+    # folder_out = os.path.join("demo_data/results/seg_maps/example-lombardia/", tileids[0])
+    folder_src = os.path.join("demo_data/lombardia/", opt.years[0], tileids[0])
+    st.markdown(""" 
+        ### Predictions
+        """)
+    # file_picker = st.selectbox("Select day predict (annual is patch-pred-nn.tif)",
+    #                            st.paths, index=st.paths.index('patch-pred-nn.tif'))
 
-    file_path = os.path.join(folder, file_picker)
+    file_path = os.path.join(folder_out, 'patch-pred-nn.tif')
     # print(file_path)
     target, profile = read(file_path)
     target = np.squeeze(target)
@@ -186,18 +256,33 @@ if st.paths is not None:
         # print(colors.to_hex(c))
         markdown_legend += f'<div style="color:gray;background-color: {colors.to_hex(c)};">{l}</div><br>'
 
-    col1, col2 = st.columns(2)
+    col1, col2 = st.columns([2,1])
     with col1:
+        st.markdown("**Long-term (annual) prediction**")
         st.pyplot(fig)
     with col2:
+        st.markdown("**Legend**")
         st.markdown(markdown_legend, unsafe_allow_html=True)
 
+    st.markdown("**Short-term (daily) prediction**")
+    img_full_paths = [os.path.join(folder_out, path) for path in st.paths if 'daily-pred' in path]
+    pred_dates = [path[:8] for path in st.paths if 'daily-pred' in path]
+    anim = load_daily_preds_as_animations(img_full_paths, pred_dates)
+    components.html(anim.to_jshtml(), height=600)
+
+    st.markdown("**Input time series**")
+    list_dir = os.listdir(folder_src)
+    list_dir.sort()
+    img_full_paths = [os.path.join(folder_src, f) for f in list_dir if f.endswith(".tif")]
+    pred_dates = [f[:8] for f in list_dir if f.endswith(".tif")]
+    anim_src = load_src_images_as_animations(img_full_paths, pred_dates)
+    components.html(anim_src.to_jshtml(), height=600)
+
 st.markdown(""" 
     ## Lombardia Dataset
     You can download other patches from the original dataset created and published on 
     [Kaggle](https://www.kaggle.com/datasets/ignazio/sentinel2-crop-mapping) and used in our paper.
-    
+
     ## How to build an input file for the Demo
-    Using a daily FPN and giving a zip that contains 30 tiff with 7 channels, correctly named you can reach prediction of 
-    crop mapping og the area...
+    Working in progress: to be defined ...
     """)