Update app.py

Add examples and information

app.py

@@ -2,6 +2,22 @@ import streamlit as st
 from transformers import AutoTokenizer, T5ForConditionalGeneration
 import post_ocr
 
+# Sidebar information
+info = '''Welcome to the demo of the [swedish-ocr-correction](https://huggingface.co/viklofg/swedish-ocr-correction) model.
+
+Enter or upload OCR output and the model will attempt to correct it.
+
+:clock2:  Slow generation? Try a shorter input.
+'''
+
+# Example inputs
+examples = {
+    'Examples': None,
+    'Example 1': 'En Gosse fur plats nu genast ! inetallyrkc, JU 83 Drottninggatan.',
+    'Example 2': '— Storartad gåfva till Göteborgs Museum. Den i HandelstidniDgens g&rdagsnnmmer omtalade hvalfisken, sorn fångats i Frölnndaviken, har i dag af hr brukspatronen James Dickson blifvit inköpt för 1,500 rdr och skänkt till härvarande Museum.',
+    'Example 3': 'Sn underlig race att ſtudera, desfa uppſinnare! utropar en Londontidnings fronifôr. Wet ni hur ſtort antalet är af patenter, ſom ſiſtlidet är utfärdades i British Patent Office? Jo, 14,000 ſty>en !! Det kan man ju fkalla en rif rd! Fjorton tuſen uppfinninnar! Herre Gud, hwilfet märkrwoärdigt tidehrvarf wi lefroa i!'
+}
+
 
 # Load model
 @st.cache_resource
@@ -23,10 +39,13 @@ post_ocr.set_model(model, tokenizer)
 
 # Title
 st.title(':memo: Swedish OCR correction')
+
+
 # Input and output areas
 tab1, tab2 = st.tabs(["Text input", "From file"])
 
 
+# Initialize session state
 def clean_inputs():
     st.session_state.inputs = {'tab1': None, 'tab2': None}
 
@@ -43,15 +62,29 @@ if 'outputs' not in st.session_state:
 
 # Sidebar (settings and stuff)
 with st.sidebar:
-    st.header('Settings')
-    n_candidates = st.number_input('Overlap', help='A higher value may lead to better quality, but takes longer time', value=1, min_value=1, max_value=7, step=2, on_change=clean_inputs)
 
-    st.header('Output')
+    st.header('Welcome')
+    st.markdown(info)
+
+    st.header('Settings')
+    overlap2candidates = {'None': 1, 'Little': 3, 'Much': 5}
+    overlap_help = '''Long texts are processed in chunks using a sliding window technique.
+    Here you can choose how much overlap the sliding window should have with the previous
+    processed chunk. No overlap is the fastest, but some overlap may increase accuracy.'''
+    overlap = st.selectbox(
+        'Overlap',
+        options=overlap2candidates,
+        help=overlap_help,
+        on_change=clean_inputs)
+    n_candidates = overlap2candidates[overlap]
+
+    st.subheader('Output')
     show_changes = st.toggle('Show changes')
 
 
 def handle_input(input_, id_):
-   
+    """Generate and display output"""
+
     with st.container(border=True):
         st.caption('Output')
 
@@ -70,8 +103,22 @@ def handle_input(input_, id_):
 
 # Manual entry tab
 with tab1:
-    typed_input = st.text_area('Input OCR', placeholder='Enter OCR generated text', label_visibility='collapsed')
-    handle_input(typed_input, 'tab1')
+    col1, col2 = st.columns([4, 1])
+
+    with col2:
+        example_title = st.selectbox('Examples', options=examples,
+                               label_visibility='collapsed')
+
+    with col1:
+        text = st.text_area(
+            label='Input text',
+            value=examples[example_titlessssssssssssssssss],
+            height=200,
+            label_visibility='collapsed',
+            placeholder='Enter OCR generated text or choose an example')
+
+        if text is not None:
+            handle_input(text, 'tab1')
 
 
 # File upload tab