update

app.py

@@ -1,53 +1,46 @@
 from fastapi import FastAPI, File, UploadFile, Form
 from fastapi.responses import JSONResponse
-from transformers import pipeline
+from transformers import BlipProcessor, TFBlipForQuestionAnswering
 from fastapi.middleware.cors import CORSMiddleware
+from PIL import Image
 
 app = FastAPI()
 
 app.add_middleware(
     CORSMiddleware,
-    allow_origins=["*"],  # Allows all origins
+    allow_origins=["*"],
     allow_credentials=True,
-    allow_methods=["*"],  # Allows all methods
-    allow_headers=["*"],  # Allows all headers
+    allow_methods=["*"],
+    allow_headers=["*"],
 )
 
-# Initialize the VQA pipeline
-vqa_pipe = pipeline("visual-question-answering", model="Salesforce/blip-vqa-capfilt-large", max_new_tokens=20)
+# Initialize the processor and model manually
+processor = BlipProcessor.from_pretrained("Salesforce/blip-vqa-capfilt-large")
+model = TFBlipForQuestionAnswering.from_pretrained("Salesforce/blip-vqa-capfilt-large")
 
 
 @app.post('/answer_question')
 async def answer_question(image: UploadFile = File(...), question: str = Form(...)):
-    """
-    This is the VQA API
-    Call this api passing an image and a question about the image
-    ---
-    parameters:
-      - name: image
-        in: formData
-        type: file
-        required: true
-      - name: question
-        in: formData
-        type: string
-        required: true
-    responses:
-      200:
-        description: Returns the answer to the question about the image
-    """
-    # Save the image locally
     image_path = 'temp_image.jpg'
     with open(image_path, 'wb') as f:
         f.write(await image.read())
 
-    # Use the VQA pipeline to get the answer
-    result = vqa_pipe(image=image_path, question=question)
+    # Open the image using PIL
+    pil_image = Image.open(image_path)
 
-    # Return the answer as JSON
-    return JSONResponse(content={'answer': result[0]['answer']})
+    # Process the image and question
+    inputs = processor(images=pil_image, text=question, return_tensors="tf")
+    pixel_values = inputs["pixel_values"]
+    input_ids = inputs["input_ids"]
+    attention_mask = inputs["attention_mask"]
+
+    outputs = model.generate(input_ids=input_ids, attention_mask=attention_mask, pixel_values=pixel_values)
+    answer = processor.decode(outputs[0], skip_special_tokens=True)
+
+    return JSONResponse(content={'answer': answer})
 
 
 if __name__ == '__main__':
     import uvicorn
+
     uvicorn.run(app, host="0.0.0.0", port=8080)