Update app.py

app.py

@@ -1,16 +1,19 @@
 import gradio as gr
 from transformers import AutoTokenizer, AutoModelForSequenceClassification
 import torch
+import easyocr
+from PIL import Image
+import numpy as np
 
-# Sarcasm detection model (public and reliable)
+# Sarcasm detection model (public, replace if needed)
 SARCASM_MODEL_NAME = "j-hartmann/emotion-english-distilroberta-base"
-sarcasm_labels = ["not sarcastic", "sarcastic"]  # simplified mapping
+sarcasm_labels = ["not sarcastic", "sarcastic"]
 
 sarcasm_tokenizer = AutoTokenizer.from_pretrained(SARCASM_MODEL_NAME)
 sarcasm_model = AutoModelForSequenceClassification.from_pretrained(SARCASM_MODEL_NAME)
 
-# Hate speech classification DeBERTa model fine-tuned for your labels (hypothetical model)
-HATE_MODEL_NAME = "your-username/deberta-hate-speech-custom"  # replace with your actual fine-tuned model
+# Hate speech classification (fine-tuned DeBERTa for your labels, replace with your actual model)
+HATE_MODEL_NAME = "your-username/deberta-hate-speech-custom"
 hate_labels = [
     "abusive_words",
     "harassment",
@@ -23,6 +26,15 @@ hate_labels = [
 hate_tokenizer = AutoTokenizer.from_pretrained(HATE_MODEL_NAME)
 hate_model = AutoModelForSequenceClassification.from_pretrained(HATE_MODEL_NAME)
 
+# OCR Reader
+reader = easyocr.Reader(['en'], gpu=False)
+
+def ocr_extract(image):
+    if isinstance(image, Image.Image):
+        image = np.array(image)
+    result = reader.readtext(image, detail=0)
+    return ' '.join(result)
+
 def detect_sarcasm(text):
     inputs = sarcasm_tokenizer(text, return_tensors="pt", truncation=True, padding=True)
     with torch.no_grad():
@@ -41,27 +53,41 @@ def classify_hate(text):
         confidence = float(probs[0][pred])
         return hate_labels[pred], confidence
 
-def chatbot(text):
-    if not text or not text.strip():
-        return "Please enter text to analyze."
+def chatbot(image=None, text=None):
+    input_text = ""
+    if image is not None:
+        input_text = ocr_extract(image)
+        if not input_text.strip():
+            return "No text found in image.", None, None
+    elif text and text.strip():
+        input_text = text
+    else:
+        return "Please provide an image or some text.", None, None
 
-    sarcasm_label, sarcasm_conf = detect_sarcasm(text)
+    sarcasm_label, sarcasm_conf = detect_sarcasm(input_text)
     if sarcasm_label == "sarcastic":
-        return f"Text is detected as SARCASTIC (Confidence: {sarcasm_conf:.2f}). Hate speech classification is skipped."
+        return f"Text is detected as SARCASTIC (Confidence: {sarcasm_conf:.2f}). Hate speech classification skipped.", None, sarcasm_label
 
-    hate_label, hate_conf = classify_hate(text)
+    hate_label, hate_conf = classify_hate(input_text)
     return (
-        f"Text is NOT sarcastic.\n"
-        f"Hate Speech Classification: {hate_label} (Confidence: {hate_conf:.2f})"
+        f"Input text: {input_text}\nHate Speech: {hate_label} (Confidence: {hate_conf:.2f})\nSarcasm: {sarcasm_label} (Confidence: {sarcasm_conf:.2f})",
+        hate_label,
+        sarcasm_label
     )
 
 iface = gr.Interface(
     fn=chatbot,
-    inputs=gr.Textbox(lines=3, placeholder="Enter text"),
-    outputs="text",
-    title="Sarcasm-aware Hate Speech Classifier",
-    description="""First detects sarcasm, and if no sarcasm, classifies hate speech into
-    detailed categories: abusive words, harassment (e.g., body shaming), religious hate, sexism, etc."""
+    inputs=[
+        gr.Image(type="pil", label="Upload Screenshot (optional)"),
+        gr.Textbox(lines=3, placeholder="Or, type/paste text here")
+    ],
+    outputs=[
+        gr.Textbox(label="Output"),
+        gr.Label(num_top_classes=len(hate_labels), label="Hate Speech Class"),
+        gr.Label(num_top_classes=2, label="Sarcasm")
+    ],
+    title="Sarcasm-aware Hate Speech Classifier with OCR",
+    description="Detect sarcasm first; if no sarcasm, classify hate speech from text or image screenshot."
 )
 
 if __name__ == "__main__":