Update requirements.txt

1.将host改为0.0.0.0，以便监听所有IP
2.修改https://github.com/Womsxd/MihoyoBBSTools/issues/198#issuecomment-2714437516等提出的时间可能为负数的情况
3.将使用resnet的推理代码整合，以精简docker镜像

README.md

@@ -14,7 +14,7 @@ api：https://github.com/ravizhan/geetest-v3-click-crack
 
 ## 运行步骤
 
-### 1.安装依赖
+### 1.安装依赖（本地必选，使用docker跳至[5-b](#docker)）
 
 （可选）a.如果要训练paddle的话还得安装paddlex及图像分类模块，安装看项目https://github.com/PaddlePaddle/PaddleX
 
@@ -24,6 +24,11 @@ api：https://github.com/ravizhan/geetest-v3-click-crack
 pip install -r requirements.txt
 ```
 
+仅推理
+```
+pip install -r requirements_without_train.txt
+```
+
 ### 2.自行准备数据集，V3和V4有区别（可选）
 
 ##### a. 训练resnet18（可选）
@@ -55,11 +60,16 @@ pip install -r requirements.txt
 - 运行 `python convert.py`（自行进去修改需要转换的模型，一般是选loss小的）
 - paddle模型转换要装paddle2onnx，详情参见https://www.paddlepaddle.org.cn/documentation/docs/guides/advanced/model_to_onnx_cn.html
 
-### 5.启动fastapi服务（必须要有训练完成的onnx格式模型）
+### 5-a.启动fastapi服务（必须要有训练完成的onnx格式模型）
 
 运行 `python main.py`（默认用的paddle的onnx模型，如果要用resnet18可以自己改注释）
 
 由于轨迹问题，可能会出现验证正确但是结果失败，所以建议增加retry次数，训练后的paddle模型正确率在99.9%以上
+### 5-b.使用docker启动服务 
+
+镜像地址为<span id="docker">luguoyixiazi/test_nine:25.3.21</span>
+
+运行时只需指定绑定的port即可
 
 ### 6.api调用
 
@@ -69,13 +79,15 @@ python调用如：
 import httpx
 
 def game_captcha(gt: str, challenge: str):
-	res = httpx.get("http://127.0.0.1:9645/pass_nine",params={'gt':gt,'challenge':challenge,'use_v3_model':True},timeout=10)
+	res = httpx.get("http://127.0.0.1:9645/pass_nine",params={'gt':gt,'challenge':challenge,'use_v3_model':True,"save_result":False},timeout=10)
 	datas = res.json()['data']
     if datas['result'] == 'success':
         return datas['validate']
     return None # 失败返回None 成功返回validate
 ```
 
+具体调用代码看使用项目，此处示例仅为API url和参数示例
+
 #### --宣传--
 
 欢迎大家支持我的其他项目喵~~~~~~~~

main.py

@@ -18,7 +18,8 @@ app = FastAPI()
 def get_pic(gt: str = Query(...), 
             challenge: str = Query(...), 
             point: str = Query(default=None), 
-            use_v3_model = Query(default=True)
+            use_v3_model = Query(default=True),
+            save_result = Query(default=False)
            ):
     print(f"开始获取:\ngt:{gt}\nchallenge:{challenge}")
     t = time.time()
@@ -47,18 +48,19 @@ def get_pic(gt: str = Query(...),
         result_list = predict_onnx(icon_image, bg_image, point)
         
     point_list = [f"{col}_{row}" for row, col in result_list]
-    wait_time = 4.0 - (time.time() - t)
+    wait_time = max(0,4.0 - (time.time() - t))
     time.sleep(wait_time)
     result = json.loads(crack.verify(point_list))
-    shutil.move(os.path.join(validate_path,pic_name),os.path.join(save_path,pic_name))
+    if save_result:
-    if 'validate' in result['data']:
+        shutil.move(os.path.join(validate_path,pic_name),os.path.join(save_path,pic_name))
-        path_2_save = os.path.join(save_pass_path,pic_name.split('.')[0])
+        if 'validate' in result['data']:
-    else:
+            path_2_save = os.path.join(save_pass_path,pic_name.split('.')[0])
-        path_2_save = os.path.join(save_fail_path,pic_name.split('.')[0])
+        else:
-    os.makedirs(path_2_save,exist_ok=True)
+            path_2_save = os.path.join(save_fail_path,pic_name.split('.')[0])
-    for pic in os.listdir(validate_path):
+        os.makedirs(path_2_save,exist_ok=True)
-        if pic.startswith('cropped'):
+        for pic in os.listdir(validate_path):
-            shutil.move(os.path.join(validate_path,pic),os.path.join(path_2_save,pic))
+            if pic.startswith('cropped'):
+                shutil.move(os.path.join(validate_path,pic),os.path.join(path_2_save,pic))
     total_time = time.time() - t
     print(f"总计耗时(含等待{wait_time}s): {total_time}\n{result}")
     return JSONResponse(content=result)
@@ -68,6 +70,6 @@ def get_pic(gt: str = Query(...),
 if __name__ == "__main__":
     from predict import predict_onnx,predict_onnx_pdl
     import uvicorn
-    print(f"{' '*10}api: http://127.0.0.1:{port}/pass_nine{' '*10}")
+    print(f"{' '*10}api: http://0.0.0.0:{port}/pass_nine{' '*10}")
     print(f"{' '*10}api所需参数：gt、challenge、point(可选){' '*10}")
-    uvicorn.run(app,port=port)
+    uvicorn.run(app,host="0.0.0.0",port=port)

predict.py

@@ -2,9 +2,9 @@ import os
 
 import numpy as np
 
-from train import MyResNet18, data_transform
+
 from crop_image import crop_image, convert_png_to_jpg,draw_points_on_image
-import torch
+
 import time
 import cv2
 from PIL import Image
@@ -13,6 +13,8 @@ import onnxruntime as ort
 
 
 def predict(icon_image, bg_image):
+    from train import MyResNet18, data_transform
+    import torch
     current_dir = os.path.dirname(os.path.abspath(__file__))
     model_path = os.path.join(current_dir, 'model', 'resnet18_38_0.021147585306924.pth')
     coordinates = [
@@ -193,9 +195,10 @@ def predict_onnx_pdl(images_path):
     target = result[-1]
     answer = [coordinates[index] for index in range(9) if result[index] == target]
     print(f"识别完成{answer}，耗时: {time.time() - start}")
-    with open(os.path.join(images_path,"nine.jpg"),'rb') as f:
+    if os.path.exists(os.path.join(images_path,"nine.jpg")):
-        bg_image = f.read()
+        with open(os.path.join(images_path,"nine.jpg"),'rb') as f:
-    draw_points_on_image(bg_image, answer)
+            bg_image = f.read()
+        draw_points_on_image(bg_image, answer)
     return answer
     
     

requirements.txt

@@ -8,4 +8,5 @@ torchvision
 Pillow
 matplotlib
 tqdm
-shutil
+uvicorn
+fastapi

requirements_without_train.txt

@@ -0,0 +1,10 @@
+httpx
+cryptography
+onnxruntime
+opencv-python
+numpy
+Pillow
+matplotlib
+tqdm
+fastapi
+uvicorn