Update app.py

app.py

@@ -1,12 +1,13 @@
 import gradio as gr
 from transformers import AutoProcessor, AutoModelForCausalLM
 import spaces
-
 import io
+import base64  # Adicionando a biblioteca base64 para decodificação
 from PIL import Image
 import subprocess
 subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
-#
+
+# Carregando o modelo e o processador
 model_id = 'J-LAB/Florence-vl3'
 model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True).to("cuda").eval()
 processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
@@ -32,22 +33,35 @@ def run_example(task_prompt, image):
     )
     return parsed_answer
 
+# Função para processar imagens, agora suportando Base64
 def process_image(image, task_prompt):
-    image = Image.fromarray(image)  # Convert NumPy array to PIL Image
+    # Verifica se a imagem é uma string base64
+    if isinstance(image, str) and image.startswith("data:image"):
+        # Extraindo a parte base64 da string
+        base64_image = image.split(",")[1]
+        # Decodificando a imagem base64
+        image = Image.open(io.BytesIO(base64.b64decode(base64_image)))
+    elif isinstance(image, bytes):
+        image = Image.open(io.BytesIO(image))
+    else:
+        image = Image.fromarray(image)  # Convertendo um array NumPy para imagem PIL, se aplicável
+
+    # Mapeando os prompts de tarefas
     if task_prompt == 'Product Caption':
         task_prompt = '<MORE_DETAILED_CAPTION>'
     elif task_prompt == 'OCR':
         task_prompt = '<OCR>'
     
+    # Chamando o exemplo com a imagem processada e o prompt da tarefa
     results = run_example(task_prompt, image)
     
-    # Remove the key and get the text value
+    # Extraindo o texto gerado a partir dos resultados
     if results and task_prompt in results:
         output_text = results[task_prompt]
     else:
         output_text = ""
 
-    # Convert newline characters to HTML line breaks
+    # Convertendo quebras de linha para quebras de linha HTML
     output_text = output_text.replace("\n\n", "<br><br>").replace("\n", "<br>")
 
     return output_text
@@ -76,16 +90,14 @@ document.querySelector('button').addEventListener('click', function() {
 });
 """
 
-single_task_list =[
-    'Product Caption', 'OCR'
-]
+single_task_list =[ 'Product Caption', 'OCR' ]
 
 with gr.Blocks(css=css) as demo:
     gr.Markdown(DESCRIPTION)
     with gr.Tab(label="Product Image Select"):
         with gr.Row():
             with gr.Column():
-                input_img = gr.Image(label="Input Picture")
+                input_img = gr.Image(label="Input Picture", tool="editor", source="upload", type="pil")  # Suporte a PIL images
                 task_prompt = gr.Dropdown(choices=single_task_list, label="Task Prompt", value="Product Caption")
                 submit_btn = gr.Button(value="Submit")
             with gr.Column():
@@ -96,20 +108,30 @@ with gr.Blocks(css=css) as demo:
         To use this model via API, you can follow the example code below:
 
         ```python
-        !pip install gradio_client
-        from gradio_client import Client, handle_file
-
-        client = Client("J-LAB/Fluxi-IA")
-        result = client.predict(
-            image=handle_file('https://raw.githubusercontent.com/gradio-app/gradio/main/test/test_files/bus.png'),
-            api_name="/process_image"
-        )
-        print(result)
+        import base64
+        from PIL import Image
+        import io
+        import requests
+
+        # Converting image to base64
+        image_path = 'path_to_image.png'
+        with open(image_path, 'rb') as image_file:
+            image_base64 = base64.b64encode(image_file.read()).decode('utf-8')
+        
+        # Preparing the payload
+        payload = {
+            "image": f"data:image/png;base64,{image_base64}",
+            "task_prompt": "Product Caption"
+        }
+
+        response = requests.post("http://your-space-url-here", json=payload)
+        print(response.json())
         ```
+
         """)
         
         submit_btn.click(process_image, [input_img, task_prompt], [output_text])
 
     demo.load(lambda: None, inputs=None, outputs=None, js=js)
 
-demo.launch(debug=True) 
+demo.launch(debug=True)