Update Human as Judge

app.py

@@ -1,119 +1,209 @@
-# import gradio as gr
-# from huggingface_hub import InferenceClient
-
-# """
-# For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-# """
-# client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
-
-
-# def respond(
-#     message,
-#     history: list[tuple[str, str]],
-#     system_message,
-#     max_tokens,
-#     temperature,
-#     top_p,
-# ):
-#     messages = [{"role": "system", "content": system_message}]
-
-#     for val in history:
-#         if val[0]:
-#             messages.append({"role": "user", "content": val[0]})
-#         if val[1]:
-#             messages.append({"role": "assistant", "content": val[1]})
-
-#     messages.append({"role": "user", "content": message})
-
-#     response = ""
-
-#     for message in client.chat_completion(
-#         messages,
-#         max_tokens=max_tokens,
-#         stream=True,
-#         temperature=temperature,
-#         top_p=top_p,
-#     ):
-#         token = message.choices[0].delta.content
-
-#         response += token
-#         yield response
-
-
-# """
-# For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-# """
-# demo = gr.ChatInterface(
-#     respond,
-#     additional_inputs=[
-#         gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-#         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-#         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-#         gr.Slider(
-#             minimum=0.1,
-#             maximum=1.0,
-#             value=0.95,
-#             step=0.05,
-#             label="Top-p (nucleus sampling)",
-#         ),
-#     ],
-# )
-
-
-# if __name__ == "__main__":
-#     demo.launch()
-
-import gradio as gr
+import os, uuid, csv, random
+from datetime import datetime
 from PIL import Image
+import gradio as gr
 
-# —— 在这里根据你自己的 Agent 框架实现这个函数 ——  
+# —— 1. 环境 & 文件准备 —— 
+os.environ["GRADIO_SSR_MODE"] = "False"    # 关掉 SSR
+
+# 确保 data 目录及子目录存在
+os.makedirs("data/images", exist_ok=True)
+
+# metadata 文件：保存每次 run 的原图、prompt、agent、结果路径
+METADATA_FILE = "data/metadata.csv"
+if not os.path.exists(METADATA_FILE):
+    with open(METADATA_FILE, "w", newline="", encoding="utf-8") as f:
+        writer = csv.DictWriter(f, fieldnames=[
+            "id","original_path","prompt",
+            "agent1","img1_path","agent2","img2_path"
+        ])
+        writer.writeheader()
+
+# evaluations 文件：保存 judge 提交的评分
+EVAL_FILE = "data/evaluations.csv"
+if not os.path.exists(EVAL_FILE):
+    with open(EVAL_FILE, "w", newline="", encoding="utf-8") as f:
+        writer = csv.DictWriter(f, fieldnames=[
+            "record_id","timestamp","task",
+            "a1_follow","a1_creativity","a1_finesse",
+            "a2_follow","a2_creativity","a2_finesse"
+        ])
+        writer.writeheader()
+
+
+# —— 2. Agent 处理 & 保存到库 —— 
 def run_agent_on_image(original_img: Image.Image, prompt: str, agent_name: str) -> Image.Image:
     """
-    调用指定 agent（模型/工具）处理图片并返回结果
-    original_img: PIL Image
-    prompt: 用户输入的描述
-    agent_name: 在下拉框里选的模型名称
+    TODO: 这里替换为你自己调用 HuggingFace API 或本地模型的逻辑
     """
-    # 示例逻辑（请替换为真正的 agent 调用）
-    # if agent_name == "Model A":
-    #     return model_a.process(original_img, prompt)
-    # elif agent_name == "Model B":
-    #     return model_b.process(original_img, prompt)
-    return original_img  # TODO: 删除这行
-
-# 可选：把可用的 agent 列表写成一个文件或者直接在这里列出
-MODEL_CHOICES = ["Model A", "Model B", "Model C"]
+    return original_img
+
+def save_to_library(orig_img, prompt, a1, a2, img1, img2):
+    """把这一组 original+prompt+两个 agent 的结果存到本地 data/ 文件夹，并在 metadata.csv 记录"""
+    rec_id = uuid.uuid4().hex
+    # 保存原图
+    orig_path = f"data/images/{rec_id}_orig.png"
+    orig_img.save(orig_path)
+    # 保存两张结果图（文件名中空格替换为下划线）
+    img1_path = f"data/images/{rec_id}_{a1.replace(' ','_')}.png"
+    img2_path = f"data/images/{rec_id}_{a2.replace(' ','_')}.png"
+    img1.save(img1_path)
+    img2.save(img2_path)
+    # 追加到 metadata.csv
+    with open(METADATA_FILE, "a", newline="", encoding="utf-8") as f:
+        writer = csv.DictWriter(f, fieldnames=[
+            "id","original_path","prompt","agent1","img1_path","agent2","img2_path"
+        ])
+        writer.writerow({
+            "id": rec_id,
+            "original_path": orig_path,
+            "prompt": prompt,
+            "agent1": a1,
+            "img1_path": img1_path,
+            "agent2": a2,
+            "img2_path": img2_path
+        })
+
+def generate_and_store(orig_img, prompt, a1, a2):
+    """处理＋保存＋返回两张结果图给 Gradio 显示"""
+    out1 = run_agent_on_image(orig_img, prompt, a1)
+    out2 = run_agent_on_image(orig_img, prompt, a2)
+    save_to_library(orig_img, prompt, a1, a2, out1, out2)
+    return out1, out2
+
+
+# —— 3. 从库中随机抽取 —— 
+def load_random_record():
+    """从 metadata.csv 随机选一条，返回 record_id、原图、prompt、两张处理图的路径"""
+    with open(METADATA_FILE, "r", encoding="utf-8") as f:
+        rows = list(csv.DictReader(f))
+    if not rows:
+        # 库空时提示
+        return "", None, "No records in library", None, None
+    rec = random.choice(rows)
+    return (
+        rec["id"],
+        rec["original_path"],
+        rec["prompt"],
+        rec["img1_path"],
+        rec["img2_path"]
+    )
 
 
+# —— 4. 保存评测结果 —— 
+def save_evaluation(record_id, task, 
+                    a1_follow, a1_creativity, a1_finesse,
+                    a2_follow, a2_creativity, a2_finesse):
+    """把打分连同 record_id 和 task 存到 evaluations.csv"""
+    with open(EVAL_FILE, "a", newline="", encoding="utf-8") as f:
+        writer = csv.DictWriter(f, fieldnames=[
+            "record_id","timestamp","task",
+            "a1_follow","a1_creativity","a1_finesse",
+            "a2_follow","a2_creativity","a2_finesse"
+        ])
+        writer.writerow({
+            "record_id": record_id,
+            "timestamp": datetime.now().isoformat(),
+            "task": task,
+            "a1_follow": a1_follow,
+            "a1_creativity": a1_creativity,
+            "a1_finesse": a1_finesse,
+            "a2_follow": a2_follow,
+            "a2_creativity": a2_creativity,
+            "a2_finesse": a2_finesse
+        })
+    return "✅ Evaluation submitted!"
+
+
+# —— 5. Gradio UI —— 
+MODEL_CHOICES = ["Model A", "Model B", "Model C"]
+TASK_CHOICES = [
+    "Image Restoration",
+    "Image Enhancement",
+    "Domain & Style Transfer",
+    "Semantic-Aware Editing",
+    "Image Composition & Expansion",
+    "Face & Appeal Editing",
+    "Steganography & Security Handling"
+]
+
 with gr.Blocks() as demo:
-    gr.Markdown("## CV Agent Arena  🎨🤖\nUpload the image you want to process, provide your requirements, select two Agents, and click 'Run Agents' to compare the results!")
-
-    with gr.Row():
-        with gr.Column():
-            original = gr.Image(type="pil", label="Upload Original Image")
-            prompt = gr.Textbox(lines=2, placeholder="e.g. ‘Make it look like a sunny day’", label="Prompt")
-        with gr.Column():
-            agent1 = gr.Dropdown(choices=MODEL_CHOICES, label="Select Agent 1")
-            agent2 = gr.Dropdown(choices=MODEL_CHOICES, label="Select Agent 2")
-
-    # 处理按钮
-    run_btn = gr.Button("Run Agents")    
-
-    with gr.Row():
-        # 左侧输出：Agent1 结果
-        with gr.Column():
-            out1 = gr.Image(type="pil", label="Agent 1 Output")
-        # 右侧输出：Agent2 结果
-        with gr.Column():
-            out2 = gr.Image(type="pil", label="Agent 2 Output")
-
-    # 按钮绑定
-    run_btn.click(
-        fn=lambda img, p, a1, a2: (run_agent_on_image(img, p, a1), run_agent_on_image(img, p, a2)),
-        inputs=[original, prompt, agent1, agent2],
-        outputs=[out1, out2],
-    )
 
-if __name__ == "__main__":
+    with gr.Tabs():
+
+        # ——— Tab 1: Agent Arena ———
+        with gr.TabItem("Agent Arena"):
+            gr.Markdown("## CV Agent Arena  🎨🤖")
+            with gr.Row():
+                with gr.Column():
+                    original = gr.Image(type="pil", label="Upload Original Image")
+                    prompt   = gr.Textbox(lines=2, label="Prompt",
+                                         placeholder="e.g. Make it look like a sunny day")
+                with gr.Column():
+                    agent1 = gr.Dropdown(choices=MODEL_CHOICES, label="Select Agent 1")
+                    agent2 = gr.Dropdown(choices=MODEL_CHOICES, label="Select Agent 2")
+            run_btn = gr.Button("Run Agents")
+            with gr.Row():
+                out1 = gr.Image(type="pil", label="Agent 1 Output")
+                out2 = gr.Image(type="pil", label="Agent 2 Output")
+
+            run_btn.click(
+                fn=generate_and_store,
+                inputs=[original, prompt, agent1, agent2],
+                outputs=[out1, out2],
+                show_api=False
+            )
+
+
+        # ——— Tab 2: Human as Judge ———
+        with gr.TabItem("Human as Judge"):
+            # 隐藏状态：保存本次抽到的 record_id
+            record_id_state = gr.State("")
+
+            task_dropdown   = gr.Dropdown(choices=TASK_CHOICES, label="Task Category")
+            judge_orig      = gr.Image(label="Original Image")
+            judge_prompt    = gr.Textbox(label="Prompt", interactive=False)
+            judge_out1      = gr.Image(label="Agent 1 Result")
+            judge_out2      = gr.Image(label="Agent 2 Result")
+
+            # 当用户选 Task（或切换到此页）时，随机抽 record
+            task_dropdown.change(
+                fn=load_random_record,
+                inputs=[],
+                outputs=[record_id_state, judge_orig, judge_prompt, judge_out1, judge_out2],
+                show_api=False
+            )
+
+            gr.Markdown("### 请对两张处理图分别打分（0–5）")
+            with gr.Row():
+                with gr.Column():
+                    gr.Markdown("#### Agent 1 Evaluation")
+                    a1_follow     = gr.Radio([0,1,2,3,4,5], label="Follow Prompt")
+                    a1_creativity = gr.Radio([0,1,2,3,4,5], label="Creativity")
+                    a1_finesse    = gr.Radio([0,1,2,3,4,5], label="Finesse/Detail")
+                with gr.Column():
+                    gr.Markdown("#### Agent 2 Evaluation")
+                    a2_follow     = gr.Radio([0,1,2,3,4,5], label="Follow Prompt")
+                    a2_creativity = gr.Radio([0,1,2,3,4,5], label="Creativity")
+                    a2_finesse    = gr.Radio([0,1,2,3,4,5], label="Finesse/Detail")
+
+            submit_btn    = gr.Button("Submit Evaluation")
+            submit_status = gr.Textbox(label="Status", interactive=False)
+
+            submit_btn.click(
+                fn=save_evaluation,
+                inputs=[
+                    record_id_state, task_dropdown,
+                    a1_follow, a1_creativity, a1_finesse,
+                    a2_follow, a2_creativity, a2_finesse
+                ],
+                outputs=[submit_status],
+                show_api=False
+            )
+
     demo.queue()
-    demo.launch()
+    demo.launch(
+        share=False,
+        show_api=False,
+        ssr_mode=False
+    )