teste

app.py

@@ -1,12 +1,12 @@
 import io
-
 import requests
 import torch
 from PIL import Image
-from rich import print
+import streamlit as st
 from transformers import AutoModelForCausalLM, AutoProcessor, GenerationConfig
 
 
+# Função para baixar e processar a imagem
 def download_image(url):
     headers = {
         'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36'
@@ -16,6 +16,7 @@ def download_image(url):
     return Image.open(io.BytesIO(resp.content)).convert("RGB")
 
 
+# Função para gerar a descrição com base na imagem e no prompt
 def generate(images, prompt, processor, model, device, dtype, generation_config):
     inputs = processor(
         images=images[:2], text=f" USER: <s>{prompt} ASSISTANT: <s>", return_tensors="pt"
@@ -25,33 +26,49 @@ def generate(images, prompt, processor, model, device, dtype, generation_config)
     return response
 
 
+# Função principal da interface Streamlit
 def main():
-    # step 1: Setup constant
+    # Definir o dispositivo de execução (GPU ou CPU)
     device = "cuda:0" if torch.cuda.is_available() else "cpu"
     dtype = torch.float16
 
-    # step 2: Load Processor and Model
-    processor = AutoProcessor.from_pretrained("StanfordAIMI/CheXagent-8b", trust_remote_code=True)
-    generation_config = GenerationConfig.from_pretrained("StanfordAIMI/CheXagent-8b")
-    model = AutoModelForCausalLM.from_pretrained(
-        "StanfordAIMI/CheXagent-8b", torch_dtype=dtype, trust_remote_code=True
-    ).to(device)
-
-    # step 3: Fetch the images
-    image_path = "https://upload.wikimedia.org/wikipedia/commons/3/3b/Pleural_effusion-Metastatic_breast_carcinoma_Case_166_%285477628658%29.jpg"
-    images = [download_image(image_path)]
-
-    # step 4: Generate the Findings section
-    for anatomy in anatomies:
-        prompt = f'Describe "{anatomy}"'
-        response = generate(images, prompt, processor, model, device, dtype, generation_config)
-        print(f"Generating the Findings for [{anatomy}]:")
-        print(response)
-
-
-if __name__ == '__main__':
-    anatomies = [
-        "Airway", "Breathing", "Cardiac", "Diaphragm",
-        "Everything else (e.g., mediastinal contours, bones, soft tissues, tubes, valves, and pacemakers)"
-    ]
-    main()
+    # Carregar o modelo e o processador
+    #processor = AutoProcessor.from_pretrained("StanfordAIMI/CheXagent-8b", trust_remote_code=True)
+    #generation_config = GenerationConfig.from_pretrained("StanfordAIMI/CheXagent-8b")
+    #model = AutoModelForCausalLM.from_pretrained(
+    #    "StanfordAIMI/CheXagent-8b", torch_dtype=dtype, trust_remote_code=True
+    #).to(device)
+
+    # Título da aplicação Streamlit
+    st.title("Análise de Imagem Médica com CheXagent")
+    
+    # Entrada para o URL da imagem
+    image_url = st.text_input("Insira o URL da imagem:", "https://upload.wikimedia.org/wikipedia/commons/3/3b/Pleural_effusion-Metastatic_breast_carcinoma_Case_166_%285477628658%29.jpg")
+    
+    # Entrada para o prompt
+    prompt = st.text_input("Insira o prompt para a análise:", "Descreva o \"Airway\"")
+    
+    # Botão para gerar a resposta
+    if st.button("Gerar Análise"):
+        try:
+            # Passo 1: Baixar a imagem
+            st.write("Baixando a imagem...")
+            images = [download_image(image_url)]
+            
+            # Exibir a imagem na interface
+            st.image(images[0], caption="Imagem fornecida", use_column_width=True)
+
+            # Passo 2: Gerar a resposta
+            st.write(f"Gerando a análise para: {prompt}...")
+            #response = generate(images, prompt, processor, model, device, dtype, generation_config)
+            
+            # Exibir a resposta gerada
+            st.subheader("Análise Gerada:")
+            #st.write(response)
+        
+        except Exception as e:
+            st.error(f"Ocorreu um erro: {e}")
+
+
+if __name__ == "__main__":
+    main()

app2.py → app_old.py

@@ -1,15 +1,21 @@
 import io
+
 import requests
 import torch
 from PIL import Image
-import streamlit as st
+from rich import print
 from transformers import AutoModelForCausalLM, AutoProcessor, GenerationConfig
 
+
 def download_image(url):
-    resp = requests.get(url)
+    headers = {
+        'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36'
+    }
+    resp = requests.get(url, headers=headers)
     resp.raise_for_status()
     return Image.open(io.BytesIO(resp.content)).convert("RGB")
 
+
 def generate(images, prompt, processor, model, device, dtype, generation_config):
     inputs = processor(
         images=images[:2], text=f" USER: <s>{prompt} ASSISTANT: <s>", return_tensors="pt"
@@ -18,31 +24,34 @@ def generate(images, prompt, processor, model, device, dtype, generation_config)
     response = processor.tokenizer.decode(output, skip_special_tokens=True)
     return response
 
+
 def main():
-    st.title("Medical Image Analysis")
-    
-    device = "cuda"
+    # step 1: Setup constant
+    device = "cuda:0" if torch.cuda.is_available() else "cpu"
     dtype = torch.float16
 
+    # step 2: Load Processor and Model
     processor = AutoProcessor.from_pretrained("StanfordAIMI/CheXagent-8b", trust_remote_code=True)
     generation_config = GenerationConfig.from_pretrained("StanfordAIMI/CheXagent-8b")
     model = AutoModelForCausalLM.from_pretrained(
         "StanfordAIMI/CheXagent-8b", torch_dtype=dtype, trust_remote_code=True
     ).to(device)
 
+    # step 3: Fetch the images
     image_path = "https://upload.wikimedia.org/wikipedia/commons/3/3b/Pleural_effusion-Metastatic_breast_carcinoma_Case_166_%285477628658%29.jpg"
     images = [download_image(image_path)]
 
-    anatomies = [
-        "Airway", "Breathing", "Cardiac", "Diaphragm",
-        "Everything else (e.g., mediastinal contours, bones, soft tissues, tubes, valves, and pacemakers)"
-    ]
-
+    # step 4: Generate the Findings section
     for anatomy in anatomies:
         prompt = f'Describe "{anatomy}"'
         response = generate(images, prompt, processor, model, device, dtype, generation_config)
-        st.subheader(f"Findings for [{anatomy}]:")
-        st.write(response)
+        print(f"Generating the Findings for [{anatomy}]:")
+        print(response)
+
 
 if __name__ == '__main__':
-    main()
+    anatomies = [
+        "Airway", "Breathing", "Cardiac", "Diaphragm",
+        "Everything else (e.g., mediastinal contours, bones, soft tissues, tubes, valves, and pacemakers)"
+    ]
+    main()