Update app.py

app.py

@@ -1,15 +1,15 @@
-# ===============================
-# Mumbai Bird Call Identifier
-# FINAL PRODUCTION app.py
-# ===============================
+# ==========================================
+# Mumbai Bird Call Identifier — FINAL FIX
+# ==========================================
 
 import streamlit as st
 import torch
 import torchaudio
 import numpy as np
+import io
+import matplotlib.cm as cm
 from torchvision import models, transforms
 from PIL import Image
-import io
 
 # ================== PAGE CONFIG ==================
 st.set_page_config(
@@ -26,6 +26,9 @@ TARGET_SAMPLES = SAMPLE_RATE * DURATION
 HIGH_CONF = 0.60
 MEDIUM_CONF = 0.35
 
+# ⚠️ MUST MATCH TRAINING
+SPECTROGRAM_COLORMAP = "magma"   # ← change ONLY if training used something else
+
 # ================== LOAD MODEL ==================
 @st.cache_resource
 def load_model():
@@ -77,27 +80,21 @@ val_transform = transforms.Compose([
     )
 ])
 
-# ================== UI HEADER ==================
+# ================== UI ==================
 st.title("🐦 Mumbai Balcony Bird Call Identifier")
 st.markdown(
     """
-Identify **204 Indian bird species** using their calls.  
-Trained on **real Mumbai & Maharashtra recordings**.
-
-📌 **Best results:**  
-• Record early morning / evening  
-• Minimal background noise  
-• 5–10 seconds duration
+Identify **204 Indian bird species** from their calls.  
+Model trained on **PNG spectrogram images** from Mumbai & Maharashtra.
 """
 )
 
-# ================== FILE UPLOAD ==================
 audio_file = st.file_uploader(
     "Upload bird call audio (WAV / MP3 / M4A / OGG)",
     type=["wav", "mp3", "m4a", "ogg"]
 )
 
-# ================== MAIN LOGIC ==================
+# ================== PROCESS ==================
 if audio_file:
     st.audio(audio_file)
 
@@ -131,11 +128,15 @@ if audio_file:
         mel = db_transform(mel)
         mel = mel.squeeze(0)
 
-        # -------- MEL → IMAGE (PNG STYLE) --------
+        # -------- MEL → COLORED PNG (CRITICAL FIX) --------
         mel_np = mel.numpy()
-        mel_img = (mel_np - mel_np.min()) / (mel_np.ptp() + 1e-8)
-        mel_img = (mel_img * 255).astype(np.uint8)
-        mel_pil = Image.fromarray(mel_img).convert("RGB")
+        mel_norm = (mel_np - mel_np.min()) / (mel_np.ptp() + 1e-8)
+
+        cmap = cm.get_cmap(SPECTROGRAM_COLORMAP)
+        colored = cmap(mel_norm)[:, :, :3]  # drop alpha
+        mel_img = (colored * 255).astype(np.uint8)
+
+        mel_pil = Image.fromarray(mel_img)
 
         # -------- MODEL INPUT --------
         model_input = val_transform(mel_pil).unsqueeze(0)
@@ -153,80 +154,32 @@ if audio_file:
     top1_prob = top5_probs[0].item()
     top1_species = class_names[top5_idx[0]]
 
-    # -------- CONFIDENCE STATUS --------
     if top1_prob >= HIGH_CONF:
         st.success("✅ High confidence identification")
-        confidence_label = "High"
     elif top1_prob >= MEDIUM_CONF:
         st.warning("⚠️ Medium confidence identification")
-        confidence_label = "Medium"
     else:
         st.error("❓ Low confidence – possibly unknown species")
-        confidence_label = "Low"
 
     st.markdown(f"## 🐦 {top1_species}")
+    st.metric("Confidence", f"{top1_prob*100:.1f}%")
 
-    st.metric(
-        label="Confidence",
-        value=f"{top1_prob*100:.1f}%",
-        delta=confidence_label
-    )
-
-    # -------- TOP 5 --------
     st.markdown("### 🔍 Other possible matches")
     for i in range(1, 5):
         st.markdown(
             f"- **{class_names[top5_idx[i]]}** — {top5_probs[i].item():.1%}"
         )
 
-    # -------- LOW CONFIDENCE EXPLANATION --------
-    if top1_prob < MEDIUM_CONF:
-        st.info(
-            """
-This recording may contain:
-- A species not in the dataset
-- Juvenile or alarm calls
-- Multiple birds calling
-- Background noise
-
-Try recording again in a quieter environment.
-"""
-        )
-
-    # -------- SPECTROGRAM --------
     st.markdown("---")
-    st.subheader("📊 Mel Spectrogram Used by the Model")
-    st.image(
-        mel_img,
-        caption="Frequency (vertical) vs Time (horizontal)",
-        use_container_width=True
-    )
-
-    # -------- SPECIES INFO CARD --------
-    with st.expander("📖 About this bird"):
-        st.markdown(
-            f"""
-**Scientific name:** {top1_species}  
-**Region:** Mumbai & Maharashtra  
-**Typical habitat:** Urban areas, forest edges, wetlands  
-**Best time to hear:** Early morning or dusk
-"""
-        )
+    st.subheader("📊 Spectrogram used by the model")
+    st.image(mel_img, use_container_width=True)
 
 else:
     st.info("👆 Upload a bird call audio file to begin")
 
-    col1, col2, col3 = st.columns(3)
-    col1.metric("Species Covered", "204")
-    col2.metric("Training Calls", "8000+")
-    col3.metric("Region", "Mumbai & MH")
-
 # ================== FOOTER ==================
 st.markdown("---")
 st.caption(
-    "⚠️ AI predictions are probabilistic. "
-    "For critical identifications, verify using field guides or experts."
-)
-st.caption(
-    "Model trained on real bird call recordings sourced from citizen science datasets."
+    "⚠️ This model predicts among known species only. "
+    "Low confidence may indicate an unseen species or noisy audio."
 )