Rubaha's picture
Update app.py
0c291d3 verified
import gradio as gr
from transformers import pipeline
from PIL import Image
import torch
from diffusers import StableDiffusionPipeline
import tempfile
from groq import Groq
import os # Replaced google.colab with os for environment variable access
import nltk
from nltk.translate.bleu_score import sentence_bleu
import json
import time
# Download NLTK data for BLEU
nltk.download('punkt')
# Initialize Groq client
client = Groq(api_key=os.getenv('GROQ_API_KEY')) # Updated to use os.getenv
# Load models
captioner = pipeline("image-to-text", model="Salesforce/blip-image-captioning-base")
pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", torch_dtype=torch.float16)
pipe = pipe.to("cuda" if torch.cuda.is_available() else "cpu")
pipe.enable_attention_slicing()
# Caching for performance
caption_cache = {}
qa_cache = {}
history = [] # Global history for report
def generate_caption(image, progress=gr.Progress()):
try:
if image is None:
return "Please upload an image.", {}
progress(0.2, "Processing image...")
pil_image = Image.open(image) if isinstance(image, str) else image
cache_key = hash(pil_image.tobytes())
if cache_key in caption_cache:
return caption_cache[cache_key], {}
caption = captioner(pil_image)[0]['generated_text']
enhanced_caption = f"A creative take: {caption}."
metrics = {"length": len(enhanced_caption.split()), "unique_words": len(set(enhanced_caption.split()))}
caption_cache[cache_key] = enhanced_caption
history.append({"action": "caption", "time": time.time()})
progress(1.0, "Caption generated!")
return enhanced_caption, metrics
except Exception as e:
return f"Error: {str(e)}", {}
def generate_image_from_caption(caption, progress=gr.Progress()):
try:
progress(0.1, "Refining prompt...")
image = pipe(caption, num_inference_steps=25, guidance_scale=7.5).images[0]
progress(0.8, "Generating image...")
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.png')
image.save(temp_file.name)
history.append({"action": "image_gen", "time": time.time()})
progress(1.0, "Image ready for download!")
return image, temp_file.name
except Exception as e:
return None, f"Error: {str(e)}"
def answer_question(image, question, progress=gr.Progress()):
try:
if not question.strip():
return "Please enter a question.", {}
progress(0.2, "Analyzing context...")
start_time = time.time()
context = ""
if image is not None:
pil_image = Image.open(image) if isinstance(image, str) else image
caption_result = captioner(pil_image)[0]['generated_text']
context = f"Based on the image description: '{caption_result}'. "
cache_key = (context, question)
if cache_key in qa_cache:
return qa_cache[cache_key], {}
prompt = f"{context}Question: {question}\nAnswer:"
progress(0.5, "Querying AI...")
chat_completion = client.chat.completions.create(
messages=[{"role": "user", "content": prompt}],
model="llama-3.1-8b-instant",
)
answer = chat_completion.choices[0].message.content.strip()
response_time = time.time() - start_time
metrics = {"response_time": response_time, "length": len(answer.split())}
qa_cache[cache_key] = answer
history.append({"action": "qa", "time": time.time()})
progress(1.0, "Answer ready!")
return answer, metrics
except Exception as e:
return f"Error: {str(e)}", {}
def evaluate_caption(caption, reference="A sample reference caption for evaluation."):
try:
if not caption:
return "No caption to evaluate."
reference_tokens = nltk.word_tokenize(reference.lower())
candidate_tokens = nltk.word_tokenize(caption.lower())
bleu = sentence_bleu([reference_tokens], candidate_tokens)
return f"BLEU Score: {bleu:.2f}, Length: {len(candidate_tokens)} words"
except Exception as e:
return f"Error: {str(e)}"
def batch_caption(images):
try:
results = []
for img_path in images:
if img_path:
pil_image = Image.open(img_path)
caption = captioner(pil_image)[0]['generated_text']
results.append(f"Image: {caption}")
history.append({"action": "batch_caption", "time": time.time()})
return "\n".join(results)
except Exception as e:
return f"Error: {str(e)}"
def generate_report():
try:
total_interactions = len(history)
avg_response_time = sum(h.get("response_time", 0) for h in history) / total_interactions if total_interactions > 0 else 0
report = {
"total_interactions": total_interactions,
"average_response_time": avg_response_time,
"actions": [h["action"] for h in history]
}
return json.dumps(report, indent=2)
except Exception as e:
return f"Error generating report: {str(e)}"
# Gradio UI with enhancements
with gr.Blocks(title="ColabCraft: Advanced AI Image Assistant", theme=gr.themes.Soft()) as demo:
gr.Markdown("""
# 🧠 ColabCraft: Advanced AI Image Assistant
**A Multimodal GenAI Project for Image Captioning, Q&A, and Generation**
*Upload images, generate captions, ask questions, create images, and evaluate results. Built with Hugging Face, Stable Diffusion, and Groq's Llama 3.1 8B.*
**Ethical Note:** This tool promotes positive AI use. Avoid uploading sensitive images. Citations: BLIP (Salesforce), Stable Diffusion (CompVis), Llama (Meta via Groq).
""")
# Shared image input
image_input = gr.Image(type="pil", label="Upload Image (Shared for Captioning & Q&A)", elem_id="upload_img")
with gr.Tabs():
with gr.TabItem("πŸ“Έ Image Captioning", elem_id="caption_tab"):
gr.Markdown("### Generate Creative Captions from Images")
with gr.Row():
with gr.Column():
caption_output = gr.Textbox(label="Generated Caption", interactive=False)
metrics_output = gr.JSON(label="Metrics")
generate_btn = gr.Button("πŸš€ Generate Caption", variant="primary")
generate_btn.click(generate_caption, inputs=image_input, outputs=[caption_output, metrics_output])
# Removed the problematic gr.Examples line
with gr.TabItem("❓ Q&A Assistant", elem_id="qa_tab"):
gr.Markdown("### Ask Questions About Images or General Topics")
with gr.Row():
question_input = gr.Textbox(label="Enter Question", placeholder="e.g., What is in the image?")
answer_output = gr.Textbox(label="AI Answer", interactive=False)
qa_metrics = gr.JSON(label="Metrics")
ask_btn = gr.Button("πŸ” Get Answer", variant="primary")
ask_btn.click(answer_question, inputs=[image_input, question_input], outputs=[answer_output, qa_metrics])
with gr.TabItem("🎨 Image Generation", elem_id="gen_tab"):
gr.Markdown("### Create Images from Text Captions")
with gr.Row():
text_input = gr.Textbox(label="Enter Caption for Generation", placeholder="e.g., A sunny beach with palm trees")
image_output = gr.Image(label="Generated Image")
download_file = gr.File(label="πŸ“₯ Download Image")
generate_img_btn = gr.Button("πŸ–ΌοΈ Generate Image", variant="primary")
generate_img_btn.click(generate_image_from_caption, inputs=text_input, outputs=[image_output, download_file])
with gr.TabItem("πŸ“Š Evaluation & Batch", elem_id="eval_tab"):
gr.Markdown("### Evaluate Captions and Process Batches")
with gr.Row():
eval_caption_input = gr.Textbox(label="Caption to Evaluate")
eval_output = gr.Textbox(label="Evaluation Results", interactive=False)
eval_btn = gr.Button("πŸ“ˆ Evaluate")
eval_btn.click(evaluate_caption, inputs=eval_caption_input, outputs=eval_output)
gr.Markdown("### Batch Captioning")
batch_input = gr.File(file_count="multiple", label="Upload Multiple Images")
batch_output = gr.Textbox(label="Batch Results", interactive=False, lines=10)
batch_btn = gr.Button("πŸ”„ Process Batch")
batch_btn.click(batch_caption, inputs=batch_input, outputs=batch_output)
with gr.TabItem("πŸ“‹ Report & Help", elem_id="report_tab"):
gr.Markdown("### Project Report & Help")
report_output = gr.Textbox(label="Generated Report", interactive=False, lines=10)
report_btn = gr.Button("πŸ“„ Generate Report")
report_btn.click(generate_report, inputs=[], outputs=report_output)
gr.Markdown("""
**Help & Features:**
- **Captioning:** Uses BLIP for accurate descriptions.
- **Q&A:** Powered by Llama 3.1 8B via Groq for contextual answers.
- **Generation:** Stable Diffusion for high-quality images.
- **Evaluation:** BLEU scores for caption quality.
- **Batch:** Process multiple images at once.
- **Report:** Summarizes usage metrics.
**For Submission:** Export notebook as PDF. Include demo video and metrics in report.
""")
if __name__ == "__main__":
demo.launch(share=True)