update app.py to use st.cache_resource (much faster/more responsive)

video_semantic_search/app.py

@@ -9,6 +9,7 @@ import pandas as pd
 import streamlit as st
 
 from pipeline import clip_wrapper
+from pipeline.process_videos import DATAFRAME_PATH
 
 NUM_FRAMES_TO_RETURN = 21
 
@@ -37,8 +38,9 @@ class SemanticSearcher:
         ]
 
 
-DATASET_PATH: Final[str] = os.environ.get("DATASET_PATH", "data/dataset.parquet")
-SEARCHER: Final[SemanticSearcher] = SemanticSearcher(pd.read_parquet(DATASET_PATH))
+@st.cache_resource
+def get_semantic_searcher():
+    return SemanticSearcher(pd.read_parquet(DATAFRAME_PATH))
 
 
 @dataclass
@@ -93,7 +95,7 @@ def display_search_results(results: list[SearchResult]) -> None:
             st.markdown(
                 f"""
                 <a href="{get_video_url(result.video_id, result.timestamp)}">
-                <img src="data:image/jpeg;base64,{result.base64_image}" alt="frame {result.frame_idx} timestamp {int(result.timestamp)}" width="100%">
+                <img src="data:image/jpeg;base64,{result.base64_image.decode()}" alt="frame {result.frame_idx} timestamp {int(result.timestamp)}" width="100%">
                 </a>
                 """,
                 unsafe_allow_html=True,
@@ -110,9 +112,10 @@ def main():
     st.markdown("_App by Ben Tenmann and Sidney Radcliffe_")
     st.text_input("What are you looking for?", key="query")
     query = st.session_state["query"]
+    searcher = get_semantic_searcher()
     if query:
         st.text("Click image to open video")
-        display_search_results(SEARCHER.search(query))
+        display_search_results(searcher.search(query))
 
 
 if __name__ == "__main__":