Update app.py

app.py

@@ -1,4 +1,4 @@
-# app.py (for a Hugging Face Space using Gradio)
+# app.py (for your Hugging Face Space/Model Repo: Hajorda/keduClassifier)
 import gradio as gr
 import torch
 import pytorch_lightning as pl
@@ -12,26 +12,28 @@ import pickle
 from PIL import Image
 import numpy as np
 import os
-import requests # For fetching funny cat GIFs
+# import requests # Commenting out as Giphy API key is not used by default
+import random # For random choice of keywords if you enable Giphy later
 from huggingface_hub import hf_hub_download
 
-# --- Re-use your model definition and loading functions ---
-# (This part would be similar to your inference.py)
-
-HF_USERNAME = "Hajorda"  # Or the username of the model owner
-HF_MODEL_NAME = "keduClasifier"
+# --- Model and Repository Configuration ---
+# This should exactly match your model repository on Hugging Face
+HF_USERNAME = "Hajorda"
+HF_MODEL_NAME = "keduClassifier" # CORRECTED: Matches your repo name
 REPO_ID = f"{HF_USERNAME}/{HF_MODEL_NAME}"
 
+# --- Inference Configuration ---
 cfg_dict_for_inference = {
-    'model_name': 'swin_tiny_patch4_window7_224', # Match training
-    'dropout_backbone': 0.1, # Match training
-    'dropout_fc': 0.2,       # Match training
+    'model_name': 'swin_tiny_patch4_window7_224', # Should match your trained model
+    'dropout_backbone': 0.1, # Should match your trained model
+    'dropout_fc': 0.2,       # Should match your trained model
     'img_size': (224, 224),
-    'num_classes': 37, # IMPORTANT: This must be correct for your trained model
+    'num_classes': 37, # This MUST match the number of classes your model was trained on
 }
 cfg_inference = Box(cfg_dict_for_inference)
 
-class PetBreedModel(pl.LightningModule): # Paste your PetBreedModel class here
+# --- PyTorch Lightning Model Definition ---
+class PetBreedModel(pl.LightningModule):
     def __init__(self, cfg: Box):
         super().__init__()
         self.cfg = cfg
@@ -39,67 +41,87 @@ class PetBreedModel(pl.LightningModule): # Paste your PetBreedModel class here
             self.cfg.model_name, pretrained=False, num_classes=0,
             in_chans=3, drop_rate=self.cfg.dropout_backbone
         )
-        h, w = self.cfg.img_size
+        # Ensure img_size is a tuple for unpacking
+        h, w = self.cfg.img_size if isinstance(self.cfg.img_size, tuple) else (224, 224)
         dummy_input = torch.randn(1, 3, h, w)
-        with torch.no_grad(): num_features = self.backbone(dummy_input).shape[-1]
+        with torch.no_grad():
+            num_features = self.backbone(dummy_input).shape[-1]
         self.fc = nn.Sequential(
             nn.Linear(num_features, num_features // 2), nn.ReLU(),
             nn.Dropout(self.cfg.dropout_fc),
             nn.Linear(num_features // 2, self.cfg.num_classes)
         )
     def forward(self, x):
-        features = self.backbone(x); output = self.fc(features)
+        features = self.backbone(x)
+        output = self.fc(features)
         return output
 
+# --- Helper Functions to Load Assets from Hugging Face Hub ---
 def load_model_from_hf_for_space(repo_id=REPO_ID, ckpt_filename="pytorch_model.ckpt"):
     model_path = hf_hub_download(repo_id=repo_id, filename=ckpt_filename)
-    # Important: Ensure cfg_inference is correctly defined with num_classes
-    if cfg_inference.num_classes is None:
+    if cfg_inference.num_classes is None: # Should be set by cfg_dict_for_inference
          raise ValueError("num_classes must be set in cfg_inference to load the model for Gradio.")
+    # Pass the cfg for the model structure
     loaded_model = PetBreedModel.load_from_checkpoint(model_path, cfg=cfg_inference, strict=False)
     loaded_model.eval()
     return loaded_model
 
 def load_label_encoder_from_hf_for_space(repo_id=REPO_ID, le_filename="label_encoder.pkl"):
     le_path = hf_hub_download(repo_id=repo_id, filename=le_filename)
-    with open(le_path, 'rb') as f: label_encoder = pickle.load(f)
+    with open(le_path, 'rb') as f:
+        label_encoder = pickle.load(f)
     return label_encoder
 
-# Load model and encoder once when the app starts
-model = load_model_from_hf_for_space()
-label_encoder = load_label_encoder_from_hf_for_space()
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model.to(device)
-
-# --- Funny elements ---
-funny_cat_keywords = ["funny cat", "silly cat", "cat meme", "derp cat"]
-GIPHY_API_KEY = "YOUR_GIPHY_API_KEY" # Optional: For more variety, get a Giphy API key
+# --- Load Model and Label Encoder (once at app startup) ---
+print(f"Loading model and label encoder from repository: {REPO_ID}")
+try:
+    model = load_model_from_hf_for_space()
+    label_encoder = load_label_encoder_from_hf_for_space()
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    model.to(device)
+    print(f"Model and label encoder loaded successfully. Using device: {device}")
+except Exception as e:
+    print(f"Error loading model or label encoder: {e}")
+    # If loading fails, the Gradio app might not work.
+    # Consider how to handle this, e.g., display an error in the UI.
+    model = None
+    label_encoder = None
+    device = "cpu"
+
+
+# --- Funny GIF Logic ---
+# funny_cat_keywords = ["funny cat", "silly cat", "cat meme", "derp cat"]
+# GIPHY_API_KEY = "YOUR_GIPHY_API_KEY" # Optional
 
 def get_funny_cat_gif(breed_name):
-    try:
-        # Use a public API if you don't have a Giphy key, or a simpler source
-        # For example, a predefined list of GIFs
-        predefined_gifs = {
-            "abyssinian": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExaWN4bDNzNWVzM2VqNHE4Ym5zN2ZzZHF0Zzh0bGRqZzRjMnhsZW5pZCZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/3oriO0OEd9QIDdllqo/giphy.gif",
-            "siamese": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExa3g0dHZtZmRncWN0cnZkNnVnMGRtYjN2ajZ2d3o1cHZtaW50ZHQ5ayZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/ICOgUNjpvO0PC/giphy.gif",
-            "default": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExNWMwNnU4NW9nZTV5c3Z0eThsOHhsOWN0Nnh0a3VzbjFxeGU0bjFuNiZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/BzyTuYCmvSORqs1ABM/giphy.gif"
-        }
-        return predefined_gifs.get(breed_name.lower().replace(" ", "_"), predefined_gifs["default"])
-
-        # If using Giphy API:
-        # search_term = f"{breed_name} {random.choice(funny_cat_keywords)}"
-        # params = {'api_key': GIPHY_API_KEY, 'q': search_term, 'limit': 1, 'rating': 'g'}
-        # response = requests.get("http://api.giphy.com/v1/gifs/search", params=params)
-        # response.raise_for_status()
-        # return response.json()['data'][0]['images']['original']['url']
-    except Exception as e:
-        print(f"Error fetching GIF: {e}")
-        return predefined_gifs["default"] # Fallback
+    # Using a predefined list for simplicity and to avoid API key requirements
+    predefined_gifs = {
+        "abyssinian": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExaWN4bDNzNWVzM2VqNHE4Ym5zN2ZzZHF0Zzh0bGRqZzRjMnhsZW5pZCZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/3oriO0OEd9QIDdllqo/giphy.gif",
+        "american bulldog": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExbHgzYXB6N3g5NThnaXU2eWR2aHljOXg3NjMzbGJwNmF6NmxkdXU2ayZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/1simplexLKhMTqI/giphy.gif", # Example for a dog breed
+        "bengal": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExbnl0Z2J6cWtub29qdjFlajQ4ZXZ6czY2ZDY0cW53b3I2amI0OHhoYSZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/BK1 SANT0sqq1q/giphy.gif",
+        "birman": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExZ3Q4NXZmMjQ1azE2dHZ2czZnNnBoNThkZ3FkY2Z0c3hqNjVqMTdhaSZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/catdogcessing/giphy.gif",
+        "bombay": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExc3N5b2c3MmgwN3JzbjRkYmdocjdhcDc3ejExZGZqZmZtbDBxdXRrcSZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/q1MeAPDDMb43K/giphy.gif",
+        "british shorthair": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExYTY3NG96bTc0bnFyOGNkaXBwcTYwdGZzZ3JwY2pscGNmbmZydG05eSZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/Lq0h93752f6J9tij39/giphy.gif",
+        "egyptian mau": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExbjZ6dmJvaDhsb3N4ZXdkOXNrbzRkYnJmMHo3MnE2bWJocjU0Mm5jayZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/3o7ZeLambpFh3TS2ZO/giphy.gif",
+        "maine coon": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExd3F6NWoyanFmY2xmcHBtMHRhMXAzaXZrYnJia3UxcDRtcXFsYjE2NSZlcD12MV9pbnRlcm5hbF_naWZfYnlfaWQmY3Q9Zw/MDrmyLuUh9A1a/giphy.gif",
+        "persian": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExYW12cDRuc3ZtZ2ZpN2Q2cjdwMHBmb2F3MzJ5d295dGRscG9hdmFpNiZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/uE4gVmbjaZmmY/giphy.gif",
+        "ragdoll": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExczZqNWs2ZWU1ZTVobXVxdTZrN2hzcGZoaDVrYnNpZGF4a3FpM3N4aCZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/ObTT5h01Xo43C/giphy.gif",
+        "russian blue": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExc3NqcHgzcnVldjA2MnQxc3oyZnp5a2R1eXZxY21hZTN4NHAwd2NyNyZlcD12MV9pbnRlcm5hbF_naWZfYnlfaWQmY3Q9Zw/114ZzmjHizvdsY/giphy.gif",
+        "siamese": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExa3g0dHZtZmRncWN0cnZkNnVnMGRtYjN2ajZ2d3o1cHZtaW50ZHQ5ayZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/ICOgUNjpvO0PC/giphy.gif",
+        "sphynx": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExcXZjdzFybXh0ZW53OHI4ZWQxazNtb3N4dDNzOGJrdmZrdXFzbnUyZSZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/mlvseq9yvZhba/giphy.gif",
+        "default": "https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExNWMwNnU4NW9nZTV5c3Z0eThsOHhsOWN0Nnh0a3VzbjFxeGU0bjFuNiZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/BzyTuYCmvSORqs1ABM/giphy.gif"
+    }
+    # Normalize breed name for lookup
+    normalized_breed_name = breed_name.lower().replace(" ", "_").replace("-", "_")
+    return predefined_gifs.get(normalized_breed_name, predefined_gifs["default"])
 
 # --- Gradio Interface Function ---
-def classify_cat_breed(image_input):
-    # Gradio provides image as a NumPy array
-    img_rgb = cv2.cvtColor(image_input, cv2.COLOR_BGR2RGB) # Ensure it's RGB if needed
+def classify_cat_breed(image_input_bgr): # Gradio image is usually BGR numpy array
+    if model is None or label_encoder is None:
+        return ("Model not loaded. Please check logs.", "Error: Model components failed to load.", "")
+
+    # Convert BGR to RGB
+    img_rgb = cv2.cvtColor(image_input_bgr, cv2.COLOR_BGR2RGB)
 
     h, w = cfg_inference.img_size
     transforms_gradio = A.Compose([
@@ -112,50 +134,66 @@ def classify_cat_breed(image_input):
     with torch.no_grad():
         logits = model(input_tensor)
         probabilities = torch.softmax(logits, dim=1)
-        # Get top N predictions if you want
-        # top_probs, top_indices = torch.topk(probabilities, 3, dim=1)
-
-        # For single prediction:
         confidence, predicted_idx = torch.max(probabilities, dim=1)
 
     predicted_breed_id = predicted_idx.item()
     predicted_breed_name = label_encoder.inverse_transform([predicted_breed_id])[0]
     conf_score = confidence.item()
 
-    # Funny message and GIF
-    funny_message = f"I'm {conf_score*100:.1f}% sure this adorable furball is a {predicted_breed_name}! What a purrfect specimen!"
-    if conf_score < 0.7:
-        funny_message += " ...Or maybe it's a new, super-rare breed only I can see. 😉"
+    funny_message = f"My AI brain (all {conf_score*100:.1f}% of it that's sure) says this purrfect creature is a **{predicted_breed_name}**!"
+    if conf_score < 0.5:
+        funny_message += " ...Though, to be honest, it could also be a very fluffy potato. My circuits are confused! 🥔"
+    elif conf_score < 0.8:
+        funny_message += " Pretty confident, but if it starts barking, don't blame me! 😜"
+    else:
+        funny_message += " Absolutely magnificent! A textbook example, if cats read textbooks. 🧐"
     
     gif_url = get_funny_cat_gif(predicted_breed_name)
 
-    # Gradio expects a dictionary for multiple outputs if you name them
-    # Or a tuple if you don't name them in gr.Interface outputs
     return (
-        f"{predicted_breed_name} (Confidence: {conf_score*100:.2f}%)",
+        f"**{predicted_breed_name.title()}** (Confidence: {conf_score*100:.2f}%)",
         funny_message,
-        gif_url # Gradio can display images/GIFs from URLs
+        gif_url
     )
 
 # --- Define the Gradio Interface ---
-title = "😸 Purrfect Breed Guesser 3000 😼"
-description = "Upload a picture of a cat, and I'll (hilariously) try to guess its breed! Powered by AI and a bit of cat-titude."
-article = "<p style='text-align: center'>Model based on Swin Transformer, fine-tuned on the Oxford-IIIT Pet Dataset. <a href='https://huggingface.co/YOUR_HF_USERNAME/my-pet-breed-classifier-swin-tiny' target='_blank'>Model Card</a></p>"
+title = "😼 KEDU's Kompletely Kooky Kat (and K9?) Klassifier! 🐶"
+description = (
+    "Upload a pic of your furry overlord (cat OR dog from the Oxford-IIIT set!), and I'll "
+    "attempt a hilariously 'accurate' breed guess. Powered by Swin Transformers and an "
+    "unhealthy obsession with pets. Results may vary, giggles guaranteed!"
+)
+# Corrected article link
+article_link_href = f"https://huggingface.co/{REPO_ID}" # Uses the correctly defined REPO_ID
+article = f"<p style='text-align: center'>Model based on Swin Transformer, fine-tuned on the Oxford-IIIT Pet Dataset. <a href='{article_link_href}' target='_blank'>Model Card & Files</a></p>"
+
+# Add some example images to your repo and reference them here
+# For example, if you add 'cat_example.jpg' and 'dog_example.jpg' to your HF repo
+example_images = [
+    ["cat_example.jpg"], # You'll need to upload this image to your HF repo
+    ["dog_example.jpg"]  # You'll need to upload this image to your HF repo
+]
+# Check if example files exist, if not, provide placeholders or skip examples
+# This check would ideally be done by trying to download them if they are remote URLs
+# For local paths in a repo, Gradio handles it if the files are present.
 
 iface = gr.Interface(
     fn=classify_cat_breed,
-    inputs=gr.Image(type="numpy", label="Upload Cat Pic! 📸"),
+    inputs=gr.Image(type="numpy", label="Upload Your Pet's Most Glamorous Shot! 📸"),
     outputs=[
-        gr.Textbox(label="🧐 My Guess Is..."),
-        gr.Textbox(label="💬 My Deep Thoughts..."),
-        gr.Image(type="filepath", label="🎉 Celebration GIF! 🎉") # 'filepath' for URLs
+        gr.Textbox(label="🧐 The AI's Verdict Is... (Breed & Confidence)"),
+        gr.Markdown(label="💬 AI's Deep (and Silly) Thoughts..."), # Markdown for bolding
+        gr.Image(type="filepath", label="🎉 Celebration/Confusion GIF! 🎉")
     ],
     title=title,
     description=description,
     article=article,
-    examples=[["example_cat1.jpg"], ["example_cat2.jpg"]], # Add paths to example images in your Space repo
-    theme=gr.themes.Soft() # Or try other themes!
+    # examples=example_images, # Uncomment if you add example images to your repo
+    theme=gr.themes.Monochrome(), # Trying a different theme
+    allow_flagging='never'
 )
 
 if __name__ == "__main__":
+    # When running locally (e.g., python app.py), this will launch the server.
+    # On Hugging Face Spaces, Spaces handles the launch.
     iface.launch()