human feedback

human_feedback_ui.py

@@ -0,0 +1,105 @@
+# To Run: uv run streamlit run human_feedback_ui.py
+import streamlit as st
+import json
+from pathlib import Path
+from PIL import Image
+import numpy as np
+
+st.set_page_config(layout="wide")
+
+data_dir = Path("data/animations")
+
+
+# List all samples
+json_files = sorted(data_dir.glob("sample-*.webp"))
+sample_names = [f.stem for f in json_files]
+
+# Find first sample without .hf.json
+def find_next_unlabeled_sample(current=None):
+    for name in sample_names:
+        hf_path = data_dir / f"{name}.hf.json"
+        if not hf_path.exists():
+            if current is None or name != current:
+                return name
+    return sample_names[0] if sample_names else None
+
+# Use session state to track current sample
+if "sample_choice" not in st.session_state:
+    st.session_state["sample_choice"] = find_next_unlabeled_sample()
+
+sample_choice = st.selectbox("Select sample", sample_names, index=sample_names.index(st.session_state["sample_choice"]) if st.session_state["sample_choice"] in sample_names else 0)
+st.session_state["sample_choice"] = sample_choice
+
+json_path = data_dir / f"{sample_choice}.json"
+webp_path = data_dir / f"{sample_choice}.webp"
+
+# Load frames
+frames = []
+if webp_path.exists():
+    im = Image.open(webp_path)
+    try:
+        while True:
+            frames.append(im.convert("RGB"))
+            im.seek(im.tell() + 1)
+    except EOFError:
+        pass
+else:
+    st.error(f"WebP file not found: {webp_path}")
+
+
+# Load scene change indices from .hf.json if exists, else from .json
+hf_path = data_dir / f"{sample_choice}.hf.json"
+if hf_path.exists():
+    with open(hf_path) as f:
+        data = json.load(f)
+    scene_change_indices = data.get("scene_change_indices", [])
+else:
+    with open(json_path) as f:
+        data = json.load(f)
+    scene_change_indices = data.get("scene_change_indices", [])
+
+st.write(f"Total frames: {len(frames)}")
+
+# Show thumbnails with selection
+selected = st.multiselect(
+    "Select transition frames (scene changes)",
+    options=list(range(len(frames))),
+    default=scene_change_indices,
+    format_func=lambda i: f"Frame {i}"
+)
+
+# Show thumbnails in a grid
+
+# Horizontal slider view for thumbnails
+
+import streamlit.components.v1 as components
+import io
+import base64
+thumb_size = 64  # Slightly larger for visibility
+slider_html = "<div style='overflow-x:auto; white-space:nowrap; padding:8px;'>"
+for i, frame in enumerate(frames):
+    buf = io.BytesIO()
+    frame.resize((thumb_size, thumb_size)).save(buf, format='PNG')
+    img_b64 = base64.b64encode(buf.getvalue()).decode('utf-8')
+    highlight = i in selected
+    border = '3px solid red' if highlight else '1px solid #ccc'
+    slider_html += f"<span style='display:inline-block; margin:2px; border:{border}; border-radius:4px;'><img src='data:image/png;base64,{img_b64}' style='width:{thumb_size}px;height:{thumb_size}px;'><div style='text-align:center;font-size:10px;color:{'red' if highlight else '#333'}'>{i}</div></span>"
+slider_html += "</div>"
+components.html(slider_html, height=thumb_size+40)
+
+# Save button
+
+
+if st.button("Save human feedback"):
+    out_path = data_dir / f"{sample_choice}.hf.json"
+    with open(out_path, "w") as f:
+        json.dump({"scene_change_indices": sorted(selected)}, f, indent=2)
+    st.success(f"Saved to {out_path}")
+    next_sample = find_next_unlabeled_sample(current=sample_choice)
+    if next_sample and next_sample != sample_choice:
+        st.info(f"Next sample available: {next_sample}")
+        if st.button("Go to next sample"):
+            st.session_state["sample_choice"] = next_sample
+            # Force rerun by updating query params (Streamlit best practice)
+            st.experimental_set_query_params(sample=next_sample)
+            st.stop()

requirements.txt

@@ -1,4 +1,5 @@
 pillow
 torch
 numpy
-matplotlib
+matplotlib
+streamlit