增加训练yolo的代码

adc.py

@@ -4,14 +4,13 @@ from maix import time
 a = adc.ADC(0, adc.RES_BIT_12)
 
 while True:
-    raw_data = a.read()
-    print(f"ADC raw data:{raw_data}")
+    # raw_data = a.read()
+    # print(f"ADC raw data:{raw_data}")
     # if raw_data > 2450:
     #     print(f"ADC raw data:{raw_data}")
     # elif raw_data < 2000:
     #     print(f"ADC raw data:{raw_data}")
-    # time.sleep_ms(50)
+    time.sleep_ms(1)
 
-    # vol = a.read_vol()
-
-    # print(f"ADC vol:{vol}")
+    vol = int(a.read_vol() * 10) / 10
+    print(f"ADC vol:{vol:.1f}, {time.time():.4f}")

app.yaml

@@ -1,11 +1,12 @@
 id: t11
 name: t11
-version: 1.2.11
+version: 1.2.12
 author: t11
 icon: ''
 desc: t11
 files:
   - 4g_download_manager.py
+  - 4g_upload_manager.py
   - app.yaml
   - archery_netcore.cpython-311-riscv64-linux-gnu.so
   - at_client.py
@@ -34,3 +35,4 @@ files:
   - vision.py
   - wifi_config_httpd.py
   - wifi.py
+  - wpa_supplicant_conf.py

design_doc/command_record.md

@@ -90,5 +90,13 @@ opencv_interactive-calibration -t=chessboard -w=9 -h=6 -sz=0.025 -v="http://192.
  D:\data\test_target_photo 是用来叠加的背景图
 
  7.1 生成靶纸及黑色三角形的截图的图片，带动动，但1.12的外框
+bak
 python .\synth_compose_yolo.py  --perspective 0.04 --perspective-prob 0.8 --color-jitter 0.6  --bg-dir D:\data\test_target_photo --fg D:\code\shooting\target_photo\write.png --out ./synth_out --class-name triangle --zip ./maix_dataset.zip --num 60 --triangles-json archery_triangles_default.json --format voc  --stage2-crop  --stage2-pad-min 0.03  --stage2-pad-max 0.18 --motion-prob 0.9 --motion-kernel-max 8 --blur-max 0 --triangle-bbox-pad-frac 0.12
 
+bak_2
+python synth_keypoints_right_angle.py   --bg-dir D:\data\test_target_photo --fg D:\code\shooting\target_photo\write.png  --triangles-json archery_triangles_default.json --out ./synth_out --num 1000  --offscreen-shift-prob 0.3  --offscreen-shift-frac 0.4   --offscreen-min-visible 1  --stage2-crop  --stage2-pad-min 0.03  --stage2-pad-max 0.18 --motion-prob 0.9 --motion-kernel-max 8 --blur-max 0   --perspective-mode planar   --yaw-max-deg 10 --pitch-max-deg 8   --roll-max-deg 4   --planar-focal-frac 1.45   --perspective-prob 0.4
+
+python synth_keypoints_right_angle.py   --bg-dir D:\data\test_target_photo --fg D:\code\shooting\target_photo\write.png  --triangles-json archery_triangles_default.json --out ./synth_out --num 1000  --offscreen-shift-prob 0.3  --offscreen-shift-frac 0.4   --offscreen-min-visible 1  --stage2-crop  --stage2-pad-min 0.03  --stage2-pad-max 0.18 --motion-prob 1.0 --motion-kernel-max 8 --blur-max 0   --perspective-mode planar   --yaw-max-deg 10 --pitch-max-deg 8   --roll-max-deg 4   --planar-focal-frac 1.45   --perspective-prob 0.4
+
+
+ python pose_pixel_metrics.py --model D:\code\archery\runs\pose\runs\pose\target_pose_train\weights\best.pt --data D:\code\archery\datasets\dataset_pose.yaml --imgsz 640

test/test_audio.py

@@ -0,0 +1,50 @@
+# test_audio.pyx
+from maix import audio, time, app, gpio
+
+def run_player_loop():
+    """
+    播放控制主循环函数
+    """
+    # 初始化音频播放器
+    p = audio.Player("/root/gun.wav")
+    p.volume(40)
+    
+    # 初始化 GPIO 引脚为输出
+    led = gpio.GPIO("A25", gpio.Mode.OUT)
+    # 设置低电平
+    led.value(0)
+    
+    # 主循环
+    while not app.need_exit():
+        led.value(1)           # 点亮 LED
+        time.sleep_ms(200)     # 保持 200ms
+        led.value(0)           # 熄灭 LED
+        p.play()               # 播放音频
+        time.sleep_ms(1000)    # 等待 1 秒
+    
+    print("play finish!")
+
+
+# 可选：添加一个简单的测试函数
+def hello():
+    return "Hello from test_audio!"
+
+
+# 可选：添加一个初始化函数
+def init_led():
+    """单独测试 GPIO"""
+    led = gpio.GPIO("A25", gpio.Mode.OUT)
+    led.value(0)
+    return "LED initialized"
+
+
+# 可选：添加一个播放测试函数
+def test_play():
+    """单独测试音频播放"""
+    p = audio.Player("/root/gun.wav")
+    p.volume(50)
+    p.play()
+    return "Playing..."
+
+
+run_player_loop()

test/test_button.py

@@ -0,0 +1,25 @@
+from maix import audio, time, app,gpio
+
+
+# button1 = gpio.GPIO("ADC", gpio.Mode.IN)
+button3 = gpio.GPIO("A26", gpio.Mode.IN) # 可用
+button2 = gpio.GPIO("A16", gpio.Mode.IN)
+#设置低电平
+from maix.peripheral import adc
+channel = 0
+res_bit = adc.RES_BIT_12
+_adc_obj = adc.ADC(channel, res_bit)
+
+
+while not app.need_exit():
+    # print(f"b1: {button1.value()}")
+
+    print(f"b2: {button2.value()}")
+
+    # print(_adc_obj.read_vol())
+    print(f"b3: {button3.value()}")
+    time.sleep_ms(50)
+
+    # time.sleep_ms(1000)
+
+

test/test_cammera.py

@@ -4,6 +4,9 @@ from maix import camera, display, time
 try:
     print("Initializing camera...")
     cam = camera.Camera(640,480)
+    # cam = camera.Camera(1280,720)
+    # cam.get_exposure_us()
+    # print("Camera exposure: ", cam.get_exposure_us())
     print("Camera initialized successfully!")
     
     disp = display.Display()

test/test_laser.py

@@ -1,172 +1,246 @@
 #!/usr/bin/env python3
 # -*- coding: utf-8 -*-
 """
-激光模块测试脚本
-用于诊断激光开关问题
-
-使用方法：
-python test_laser.py
-
-功能：
-1. 初始化串口
-2. 循环测试激光开/关
-3. 打印详细调试信息
+M01激光测距模块测试脚本 - 修正版
+基于文档中的完整命令示例
 """
 
 from maix import uart, pinmap, time
+import binascii
 
 # ==================== 配置 ====================
-UART_PORT = "/dev/ttyS1"   # 激光模块连接的串口（UART1）
-BAUDRATE = 9600            # 波特率
-
-# 引脚映射（确保与硬件连接一致）
-print("=" * 50)
-print("🔧 步骤1: 配置引脚映射")
-print("=" * 50)
+UART_PORT = "/dev/ttyS1"
+BAUDRATE = 9600
 
+# 初始化串口
 try:
     pinmap.set_pin_function("A18", "UART1_RX")
-    print("✅ A18 -> UART1_RX")
-except Exception as e:
-    print(f"❌ A18 配置失败: {e}")
-
-try:
     pinmap.set_pin_function("A19", "UART1_TX")
-    print("✅ A19 -> UART1_TX")
-except Exception as e:
-    print(f"❌ A19 配置失败: {e}")
-
-# ==================== 激光控制指令 ====================
-MODULE_ADDR = 0x00
-
-# 原始命令
-LASER_ON_CMD = bytes([0xAA, MODULE_ADDR, 0x01, 0xBE, 0x00, 0x01, 0x00, 0x01, 0xC1])
-LASER_OFF_CMD = bytes([0xAA, MODULE_ADDR, 0x01, 0xBE, 0x00, 0x01, 0x00, 0x00, 0xC0])
-
-# 备用命令格式（如果原始命令不工作，可以尝试这些）
-# 格式1: 简化命令
-LASER_ON_CMD_ALT1 = bytes([0xAA, 0x01, 0x01])
-LASER_OFF_CMD_ALT1 = bytes([0xAA, 0x01, 0x00])
-
-# 格式2: 不同的协议头
-LASER_ON_CMD_ALT2 = bytes([0x55, 0xAA, 0x01])
-LASER_OFF_CMD_ALT2 = bytes([0x55, 0xAA, 0x00])
-
-print("\n" + "=" * 50)
-print("🔧 步骤2: 初始化串口")
-print("=" * 50)
-print(f"设备: {UART_PORT}")
-print(f"波特率: {BAUDRATE}")
-
-try:
     laser_uart = uart.UART(UART_PORT, BAUDRATE)
-    print(f"✅ 串口初始化成功: {laser_uart}")
+    print("✅ 硬件初始化完成")
 except Exception as e:
-    print(f"❌ 串口初始化失败: {e}")
+    print(f"❌ 初始化失败: {e}")
     exit(1)
 
-# ==================== 测试函数 ====================
-def send_and_check(cmd, name):
-    """发送命令并检查回包"""
-    print(f"\n📤 发送: {name}")
-    print(f"   命令字节: {cmd.hex()}")
-    print(f"   命令长度: {len(cmd)} 字节")
+# ==================== 根据文档的完整命令集 ====================
+# 1. 激光开关（文档2.3.10，已验证可用）
+LASER_ON_CMD = bytes([0xAA, 0x00, 0x01, 0xBE, 0x00, 0x01, 0x00, 0x01, 0xC1])
+LASER_OFF_CMD = bytes([0xAA, 0x00, 0x01, 0xBE, 0x00, 0x01, 0x00, 0x00, 0xC0])
 
-    # 清空接收缓冲区
+# 2. 尝试不同的测距命令格式
+TEST_COMMANDS = [
+    # 格式1：文档2.3.12的单次测量（您测试失败的）
+    {
+        "name": "单次测量 (0x0020)",
+        "cmd": bytes([0xAA, 0x00, 0x00, 0x20, 0x00, 0x01, 0x00, 0x00, 0x21]),
+        "desc": "文档2.3.12 示例命令"
+    },
+    # 格式2：文档2.3.7的读取测量结果
+    {
+        "name": "读取测量结果 (0x0022)",
+        "cmd": bytes([0xAA, 0x80, 0x00, 0x22, 0xA2]),
+        "desc": "文档2.3.7 读取测量结果"
+    },
+    # 格式3：文档2.3.13的快速测量
+    {
+        "name": "快速测量 (0x0022带数据)",
+        "cmd": bytes([0xAA, 0x00, 0x00, 0x22, 0x00, 0x01, 0x00, 0x00, 0x23]),
+        "desc": "文档2.3.13 快速测量"
+    },
+    # 格式4：连续测量模式
+    {
+        "name": "连续测量模式 (0x0021)",
+        "cmd": bytes([0xAA, 0x00, 0x00, 0x21, 0x00, 0x01, 0x00, 0x00, 0x22]),
+        "desc": "文档2.3.14 连续测量"
+    }
+]
+
+def clear_buffer():
+    """清空串口缓冲区"""
     try:
-        old_data = laser_uart.read(-1)
-        if old_data:
-            print(f"   清空缓冲区: {len(old_data)} 字节")
+        data = laser_uart.read(-1)
+        if data:
+            print(f"清空: {len(data)}字节")
     except:
         pass
 
-    # 发送命令
+def send_and_wait(cmd, name, wait_time=2000):
+    """发送命令并等待响应"""
+    print(f"\n📤 发送: {name}")
+    print(f"   命令: {cmd.hex()}")
+    
+    clear_buffer()
+    
     try:
-        written = laser_uart.write(cmd)
-        print(f"   写入字节数: {written}")
+        laser_uart.write(cmd)
+        print(f"   已发送 {len(cmd)} 字节")
     except Exception as e:
-        print(f"   ❌ 写入失败: {e}")
+        print(f"   ❌ 发送失败: {e}")
         return None
     
     # 等待响应
-    time.sleep_ms(100)
+    start_time = time.ticks_ms()
+    response = b""
     
-    # 读取回包
+    while time.ticks_ms() - start_time < wait_time:
         try:
-        resp = laser_uart.read(50)
-        if resp:
-            print(f"   📥 收到回包: {resp.hex()} ({len(resp)} 字节)")
-            return resp
+            chunk = laser_uart.read(1)
+            if chunk:
+                response += chunk
+                # 完整响应通常是9或13字节
+                if len(response) >= 9:
+                    # 检查是否完整帧
+                    if response[0] in [0xAA, 0xEE]:
+                        if len(response) >= 13:  # 测距完整响应
+                            break
+                        elif response[0] == 0xEE:  # 错误响应
+                            break
+        except:
+            break
+        
+        time.sleep_ms(10)
+    
+    if response:
+        print(f"   📥 响应: {response.hex()}")
+        print(f"   长度: {len(response)} 字节")
+        
+        # 解析错误码
+        if response[0] == 0xEE and len(response) >= 9:
+            err_code = (response[7] << 8) | response[8]
+            error_mapping = {
+                0x0000: "无错误",
+                0x0001: "硬件错误",
+                0x0002: "无输出数据", 
+                0x0003: "反射信号太弱",
+                0x0004: "反射信号太强",
+                0x0005: "温度太高(>40℃)",
+                0x0006: "温度太低(<-10℃)",
+                0x0007: "电源电压低(<2.5V)",
+                0x0008: "超出量程",
+                0x0009: "读通讯错误",
+                0x000A: "写通讯错误",
+                0x000B: "地址错误"
+            }
+            err_msg = error_mapping.get(err_code, f"未知错误: 0x{err_code:04X}")
+            print(f"   ❌ 模块错误: {err_msg}")
     else:
-            print(f"   ⚠️ 无回包")
-            return None
-    except Exception as e:
-        print(f"   ❌ 读取失败: {e}")
+        print("   ⚠️ 无响应")
+    
+    return response
+
+def parse_distance_data(response):
+    """解析距离数据"""
+    if not response or len(response) < 13:
         return None
     
-def test_laser_cycle(on_cmd, off_cmd, cmd_name="标准命令"):
-    """测试一个开关周期"""
-    print(f"\n{'='*50}")
-    print(f"🧪 测试 {cmd_name}")
-    print(f"{'='*50}")
+    if response[0] != 0xAA or response[3] not in [0x20, 0x21, 0x22]:
+        return None
     
-    print("\n>>> 测试开启激光")
-    send_and_check(on_cmd, f"{cmd_name} - 开启")
-    print("   ⏱️ 等待 2 秒观察激光是否亮起...")
-    time.sleep(2)
+    # 解析4字节BCD码
+    bcd_bytes = response[6:10]
+    distance_int = 0
     
-    print("\n>>> 测试关闭激光")
-    send_and_check(off_cmd, f"{cmd_name} - 关闭")
-    print("   ⏱️ 等待 2 秒观察激光是否熄灭...")
-    time.sleep(2)
+    for byte in bcd_bytes:
+        high = (byte >> 4) & 0x0F
+        low = byte & 0x0F
+        
+        if high > 9 or low > 9:
+            return None
+        
+        distance_int = distance_int * 100 + high * 10 + low
+    
+    distance_m = distance_int / 1000.0
+    
+    # 信号质量
+    signal = 0
+    if len(response) >= 12:
+        signal = (response[10] << 8) | response[11]
+    
+    return {
+        'meters': distance_m,
+        'millimeters': distance_m * 1000,
+        'signal': signal,
+        'raw': response.hex()
+    }
 
 # ==================== 主测试 ====================
 print("\n" + "="*50)
-print("🚀 开始激光测试")
+print("M01激光测距模块详细测试")
 print("="*50)
-print("\n请观察激光模块的状态变化...")
-print("测试将依次尝试不同的命令格式\n")
 
 try:
-    # 测试1: 标准命令
-    test_laser_cycle(LASER_ON_CMD, LASER_OFF_CMD, "标准命令")
+    # 1. 测试基本连接
+    print("\n1. 测试模块连接...")
+    version_cmd = bytes([0xAA, 0x80, 0x00, 0x0A, 0x8A])
+    resp = send_and_wait(version_cmd, "读取硬件版本")
     
-    input("\n按回车继续测试备用命令1...")
+    if resp and resp[0] == 0xAA and resp[3] == 0x0A:
+        print(f"✅ 模块正常，版本: {resp[6]:02X}{resp[7]:02X}")
+    else:
+        print("❌ 模块连接测试失败")
+        exit(1)
     
-    # 测试2: 备用命令格式1
-    test_laser_cycle(LASER_ON_CMD_ALT1, LASER_OFF_CMD_ALT1, "备用命令1 (简化)")
+    # 2. 开启激光
+    print("\n2. 开启激光...")
+    resp = send_and_wait(LASER_ON_CMD, "开启激光", 1000)
+    if resp and resp.hex() == "aa0001be00010001c1":
+        print("✅ 激光已开启")
     
-    input("\n按回车继续测试备用命令2...")
+    print("   等待激光稳定...")
+    time.sleep(2)  # 重要等待时间
     
-    # 测试3: 备用命令格式2
-    test_laser_cycle(LASER_ON_CMD_ALT2, LASER_OFF_CMD_ALT2, "备用命令2 (0x55AA头)")
+    # 3. 尝试不同的测距命令
+    print("\n3. 测试不同测距命令...")
+    
+    for i, test_cmd in enumerate(TEST_COMMANDS):
+        print(f"\n{'='*30}")
+        print(f"测试 {i+1}: {test_cmd['name']}")
+        print(f"{test_cmd['desc']}")
+        print(f"{'='*30}")
+        
+        resp = send_and_wait(test_cmd['cmd'], test_cmd['name'], 3000)
+        
+        if resp:
+            if resp[0] == 0xAA and len(resp) >= 13:
+                result = parse_distance_data(resp)
+                if result:
+                    print(f"✅ 测距成功!")
+                    print(f"   距离: {result['meters']:.3f} m")
+                    print(f"   距离: {result['millimeters']:.1f} mm")
+                    print(f"   信号质量: {result['signal']}")
+                    break
+                else:
+                    print("❌ 无法解析距离数据")
+            elif resp[0] == 0xEE:
+                print("❌ 命令执行错误")
+            else:
+                print("❌ 无效响应格式")
+        else:
+            print("❌ 无响应")
+        
+        time.sleep(1)  # 命令间间隔
+    
+    # 4. 关闭激光
+    print("\n4. 关闭激光...")
+    send_and_wait(LASER_OFF_CMD, "关闭激光", 1000)
     
     print("\n" + "="*50)
     print("🏁 测试完成")
     print("="*50)
-    print("\n诊断建议：")
-    print("1. 如果激光始终不亮/始终亮：")
-    print("   - 检查激光模块的电源连接")
-    print("   - 检查串口TX/RX是否接反")
-    print("   - 尝试不同的波特率 (4800/19200)")
-    print("")
-    print("2. 如果有回包但激光无反应：")
-    print("   - 命令格式可能正确但激光硬件问题")
-    print("")
-    print("3. 如果某个备用命令有效：")
-    print("   - 需要更新 config.py 中的命令格式")
+    
+    print("\n📋 测试总结：")
+    print("1. 模块通信: ✅ 正常")
+    print("2. 激光控制: ✅ 正常")
+    print("3. 测距功能: ❌ 有问题")
+    print("\n建议：")
+    print("1. 检查激光是否实际发光（在暗处观察红点）")
+    print("2. 确保测量目标在有效范围内（0.2-60米）")
+    print("3. 确保目标有足够反射率（白色平面最佳）")
+    print("4. 如果所有测距命令都返回ERR_ADDR，可能是固件版本问题")
     
 except KeyboardInterrupt:
-    print("\n\n🛑 测试被中断")
-    # 确保激光关闭
+    print("\n\n🛑 用户中断")
     laser_uart.write(LASER_OFF_CMD)
     print("✅ 已发送关闭指令")
 except Exception as e:
     print(f"\n❌ 测试出错: {e}")
-    import traceback
-    traceback.print_exc()
-
-
-
-
-

test/test_motor.py

@@ -0,0 +1,16 @@
+from maix import gpio, pinmap, time
+ 
+
+#设置引脚为输出
+led = gpio.GPIO("A25", gpio.Mode.OUT)
+#设置低电平
+led.value(0)
+ 
+while 1:
+    # time.sleep_ms(1000)
+    #对该引脚的电平进行取反（原高-》现低）
+    # led.toggle()
+    led.value(1)
+    #延时
+    time.sleep_ms(5000)
+    led.value(0)

train_yolo/synth_compose_yolo.py

@@ -28,6 +28,11 @@ Stage2 ROI（对齐「先检整靶再裁小图」的第二步输入）：
   - --motion-prob：施加概率；--motion-kernel-min/max：模糊 streak 长度（奇数核，越大越糊）。
   - 可与 --blur-max 高斯模糊叠加；Stage2 建议：--motion-prob 0.5~0.7 --motion-kernel-max 35 --blur-max 1.2
 
+透视有两种模式（见 --perspective-mode）：
+  - corner：四角独立随机抖动（旧版），组合后易出现「不物理」的夸张梯形；--perspective 为 jitter 强度。
+  - planar：把靶当作平面，按 yaw/pitch/roll（度）随机旋转再投影，便于限定在例如「偏航 ±45°、俯仰 ±30°」内，
+    更接近 5m 外拍 40cm 靶等场景（焦距用 --planar-focal-frac 控制视场）。
+
 依赖：OpenCV + NumPy（PC 上跑即可；Maix 上若内存够也可试）。
 
 示例：
@@ -266,6 +271,74 @@ def _paste_fg_on_bg(bg_bgr, x, y, fg_scaled_bgra):
     roi_bg[:] = blended.astype(np.uint8)
 
 
+def _rotation_matrix_ypr(yaw_deg: float, pitch_deg: float, roll_deg: float, np) -> np.ndarray:
+    """靶平面绕相机坐标原点旋转：先 yaw(Y)，再 pitch(X)，再 roll(Z)。单位：度。"""
+    y, p, r = np.radians([yaw_deg, pitch_deg, roll_deg])
+    cy, sy = np.cos(y), np.sin(y)
+    cp, sp = np.cos(p), np.sin(p)
+    cr, sr = np.cos(r), np.sin(r)
+    Ry = np.array([[cy, 0.0, sy], [0.0, 1.0, 0.0], [-sy, 0.0, cy]], dtype=np.float64)
+    Rx = np.array([[1.0, 0.0, 0.0], [0.0, cp, -sp], [0.0, sp, cp]], dtype=np.float64)
+    Rz = np.array([[cr, -sr, 0.0], [sr, cr, 0.0], [0.0, 0.0, 1.0]], dtype=np.float64)
+    return Rz @ Rx @ Ry
+
+
+def _perspective_warp_planar_rgba(
+    img_bgra,
+    yaw_deg: float,
+    pitch_deg: float,
+    roll_deg: float,
+    focal_frac: float,
+    np,
+    cv2,
+):
+    """
+    平面靶透视：将 frontal 图像视作 z=1 平面上的针孔投影，再旋转三维射线方向，等价于靶相对相机倾斜。
+    focal_frac：焦距 fx=fy=max(w,h)*focal_frac，越大视场越窄、透视越温和。
+    返回 (warped BGRA, M)；退化时返回 (copy, None)。
+    """
+    h, w = img_bgra.shape[:2]
+    if min(w, h) < 16:
+        return img_bgra.copy(), None
+    fx = fy = float(max(w, h) * max(0.25, focal_frac))
+    cx, cy = w * 0.5, h * 0.5
+    R = _rotation_matrix_ypr(yaw_deg, pitch_deg, roll_deg, np)
+
+    uv_dst: list[list[float]] = []
+    z_min = 0.08
+    for u, v in ((0.0, 0.0), (float(w), 0.0), (float(w), float(h)), (0.0, float(h))):
+        x = (u - cx) / fx
+        y = (v - cy) / fy
+        z = 1.0
+        vec = R @ np.array([x, y, z], dtype=np.float64)
+        if float(vec[2]) < z_min:
+            return img_bgra.copy(), None
+        uu = fx * (vec[0] / vec[2]) + cx
+        vv = fy * (vec[1] / vec[2]) + cy
+        uv_dst.append([uu, vv])
+
+    pts_dst = np.float32(uv_dst)
+    xmin = float(pts_dst[:, 0].min())
+    ymin = float(pts_dst[:, 1].min())
+    pts_shift = pts_dst.copy()
+    pts_shift[:, 0] -= xmin
+    pts_shift[:, 1] -= ymin
+    out_w = max(4, int(np.ceil(float(pts_shift[:, 0].max()))) + 2)
+    out_h = max(4, int(np.ceil(float(pts_shift[:, 1].max()))) + 2)
+
+    pts_src = np.float32([[0, 0], [w, 0], [w, h], [0, h]])
+    M = cv2.getPerspectiveTransform(pts_src, pts_shift)
+    warped = cv2.warpPerspective(
+        img_bgra,
+        M,
+        (out_w, out_h),
+        flags=cv2.INTER_LINEAR,
+        borderMode=cv2.BORDER_CONSTANT,
+        borderValue=(0, 0, 0, 0),
+    )
+    return warped, M
+
+
 def _perspective_warp_rgba(img_bgra, jitter_frac: float, rng: random.Random, np, cv2):
     """
     对前景做轻微透视（四角微移），返回 (warped BGRA, M)。
@@ -311,6 +384,26 @@ def _perspective_warp_rgba(img_bgra, jitter_frac: float, rng: random.Random, np,
     return warped, M
 
 
+def _apply_perspective_warp_fg(fg_bgra, rng: random.Random, np, cv2, args) -> tuple:
+    """在已由 perspective_prob 抽中的样本上做一次透视。返回 (fg_out, M|None)。"""
+    mode = getattr(args, "perspective_mode", "corner")
+    if mode == "planar":
+        
+        x = rng.betavariate(2,2)
+        yaw = -float(args.yaw_max_deg) + x * (float(args.yaw_max_deg) +float(args.yaw_max_deg))
+        x = rng.betavariate(2,2)
+        pitch = -float(args.pitch_max_deg) + x * (float(args.pitch_max_deg) +float(args.pitch_max_deg))
+        x = rng.betavariate(2,2)
+        roll = -float(args.roll_max_deg) + x * (float(args.roll_max_deg) +float(args.roll_max_deg))
+     
+        focal_frac = float(getattr(args, "planar_focal_frac", 1.25))
+        return _perspective_warp_planar_rgba(fg_bgra, yaw, pitch, roll, focal_frac, np, cv2)
+    jitter = float(getattr(args, "perspective", 0.0))
+    if jitter <= 0:
+        return fg_bgra.copy(), None
+    return _perspective_warp_rgba(fg_bgra, jitter, rng, np, cv2)
+
+
 def _color_jitter_bgr(comp_bgr, strength: float, rng: random.Random, np, cv2):
     """整图 HSV 抖动：strength∈[0,1] 越大越强。"""
     if strength <= 1e-6:
@@ -519,11 +612,17 @@ def main():
         help="运动模糊 streak 长度上限，越大越像长曝光/手抖",
     )
     ap.add_argument("--jpeg-quality", type=int, default=92)
+    ap.add_argument(
+        "--perspective-mode",
+        choices=("corner", "planar"),
+        default="corner",
+        help="corner=四角随机抖动（易夸张）；planar=yaw/pitch/roll 平面投影（易限定视姿）",
+    )
     ap.add_argument(
         "--perspective",
         type=float,
         default=0.0,
-        help="轻微透视：四角扰动约为 min(靶宽,靶高)×该系数，0 关闭（建议 0.02~0.06）",
+        help="仅 perspective-mode=corner：四角扰动≈min(靶宽,靶高)×该系数，0 关闭（温和可用 0.015~0.03）",
     )
     ap.add_argument(
         "--perspective-prob",
@@ -531,6 +630,30 @@ def main():
         default=0.75,
         help="每张图应用透视的概率 0~1",
     )
+    ap.add_argument(
+        "--yaw-max-deg",
+        type=float,
+        default=45.0,
+        help="planar：偏航角均匀采样上界（度），实际 ∈[-max,max]",
+    )
+    ap.add_argument(
+        "--pitch-max-deg",
+        type=float,
+        default=30.0,
+        help="planar：俯仰角均匀采样上界（度）",
+    )
+    ap.add_argument(
+        "--roll-max-deg",
+        type=float,
+        default=8.0,
+        help="planar：滚转角均匀采样上界（度），手持可略大",
+    )
+    ap.add_argument(
+        "--planar-focal-frac",
+        type=float,
+        default=1.25,
+        help="planar：fx=fy=max(w,h)×该值，越大透视越温和（建议 1.1~1.8）",
+    )
     ap.add_argument(
         "--color-jitter",
         type=float,
@@ -662,8 +785,15 @@ def main():
         fg_s = cv2.resize(fg_crop, (new_w, new_h), interpolation=cv2.INTER_AREA)
 
         persp_M = None
-        if args.perspective > 0 and rng.random() < args.perspective_prob:
-            fg_s, persp_M = _perspective_warp_rgba(fg_s, args.perspective, rng, np, cv2)
+        want_p = (
+            rng.random() < args.perspective_prob
+            and (
+                args.perspective_mode == "planar"
+                or (args.perspective_mode == "corner" and args.perspective > 0)
+            )
+        )
+        if want_p:
+            fg_s, persp_M = _apply_perspective_warp_fg(fg_s, rng, np, cv2, args)
 
         fw2, fh2 = fg_s.shape[1], fg_s.shape[0]
         tx0, ty0, tw, th = _fg_bbox_from_alpha(fg_s)

train_yolo/train_black_triangle_pos.py

@@ -0,0 +1,506 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+YOLO11 关键点检测训练脚本（靶纸四角）。
+
+设备优先级（--device auto）：Intel XPU > NVIDIA CUDA > CPU。
+默认 imgsz=960；批大小默认 4（大图显存紧张时可再降）。
+
+关于「业务像素误差」：
+  Ultralytics 没有在 yaml 里设定「像素阈值」的选项；反向传播仍由 pose/kobj/box 等内部 loss 驱动。
+  - 监控：--pixel-metrics-every N（每 N 个 epoch 打印 mean/p95，并合并进 runs/.../results.csv，见 pose_pixel_metrics.py）。
+  - 选 best.pt / early stopping：加 --best-by-pixel，用验证集 mean 像素误差（与 pose_pixel_metrics
+    同一口径）代替 mAP 合成 fitness（fitness = -mean_px，越小越好）。
+    多卡 DDP（world_size>1）时会自动退回默认 mAP fitness。
+
+  XPU：Ultralytics BaseTrainer._get_memory / _clear_memory 把非 MPS、非 CPU 一律当 CUDA，
+  会在验证前调用 torch.cuda 而报错；本脚本在选用 XPU 时自动打补丁（见 _patch_ultralytics_trainer_for_xpu）。
+
+  务必使用 pose 任务：YOLO(...) 与 model.train(...) 均指定 task='pose'。若误用默认 detect，
+  会把 17 列 Pose 标注当成检测/分割解析，校验时出现「coordinates > 1」或 [2.] 等假象。
+"""
+
+from __future__ import annotations
+
+import argparse
+import csv
+import gc
+import glob
+import os
+import tempfile
+from copy import deepcopy
+from pathlib import Path
+
+import torch
+from ultralytics import YOLO
+
+from pose_pixel_metrics import eval_val_pixel_error
+import warnings
+warnings.filterwarnings('ignore', 
+    message=".*scatter_add_kernel does not have a deterministic implementation.*")
+
+
+def _clear_ultralytics_label_caches(data_yaml_path: str) -> int:
+    """删除 data.yaml 的 path 下 labels/*.cache。
+
+    Ultralytics 的校验缓存 hash 仅依赖「标签/图片路径字符串 + 各文件 size 之和」，不含文件内容；
+    修正 *.txt 后若总和巧合不变，可能继续加载旧 cache 并重播旧的 corrupt 日志，训练前应删掉。"""
+    from ultralytics.utils import YAML
+
+    try:
+        cfg = YAML.load(data_yaml_path)
+    except Exception:
+        return 0
+    root = cfg.get("path")
+    if not root:
+        return 0
+    root = os.path.abspath(os.path.expanduser(str(root)))
+    pattern = os.path.join(root, "labels", "*.cache")
+    n = 0
+    for p in glob.glob(pattern):
+        try:
+            os.unlink(p)
+            n += 1
+        except OSError:
+            pass
+    return n
+
+
+def _pick_device(explicit: str | None):
+    """返回 ultralytics train/predict 可用的 device。"""
+    if explicit and explicit != "auto":
+        e = explicit.lower()
+        if e == "xpu":
+            if getattr(torch, "xpu", None) is None or not torch.xpu.is_available():
+                raise RuntimeError("指定了 --device xpu 但当前环境不可用")
+            return torch.device("xpu")
+        if e in ("0", "cuda", "gpu"):
+            if not torch.cuda.is_available():
+                raise RuntimeError("指定了 CUDA 但不可用")
+            return 0
+        if e == "cpu":
+            return "cpu"
+        return explicit
+    if getattr(torch, "xpu", None) is not None and torch.xpu.is_available():
+        return torch.device("xpu")
+    if torch.cuda.is_available():
+        return 0
+    return "cpu"
+
+
+def _default_amp(device) -> bool:
+    if isinstance(device, torch.device) and device.type == "xpu":
+        return False
+    if device == "cpu":
+        return False
+    return True
+
+
+def _patch_ultralytics_for_xpu():
+    """为 Ultralytics 打补丁，使其能在 XPU 环境下正常训练和验证。"""
+    import ultralytics.engine.trainer as ut_trainer
+    import ultralytics.engine.validator as ut_validator
+    from ultralytics.utils.torch_utils import select_device as _original_select_device
+
+    # 1. 覆盖 select_device：Trainer 初始化传入 torch.device("xpu") 会走原版早返回；
+    #    初始化后 args.device 会变成字符串 "xpu"，中期 val 用 trainer.device，不调用 select_device；
+    #    训练结束 final_eval 里 Validator 会 select_device("xpu")，且 validator 在 import 时已绑定原函数，
+    #    只改 torch_utils 无效，必须同时修补 trainer/validator 模块内的引用。
+    def _patched_select_device(device="", *args, **kwargs):
+        # Ultralytics 8.4.x: select_device(device="", newline=False, verbose=True)
+        # Older forks sometimes passed extra positional args; forward everything.
+        if isinstance(device, str):
+            d = device.strip().lower()
+            if d == "xpu" or d.startswith("xpu:"):
+                return torch.device(device.strip())
+        return _original_select_device(device, *args, **kwargs)
+
+    import ultralytics.utils.torch_utils
+
+    ultralytics.utils.torch_utils.select_device = _patched_select_device
+    ut_trainer.select_device = _patched_select_device
+    ut_validator.select_device = _patched_select_device
+
+    # 2. 修补 Trainer 的内存函数
+    BT = ut_trainer.BaseTrainer
+    if not getattr(BT, "_archery_xpu_memory_patched", False):
+        _orig_get_memory = BT._get_memory
+        _orig_clear_memory = BT._clear_memory
+
+        def _get_memory(self, fraction=False):
+            if self.device.type != "xpu":
+                return _orig_get_memory(self, fraction)
+            # ... (原有的 XPU 内存获取逻辑保持不变) ...
+            memory, total = 0, 0
+            try:
+                idx = self.device.index
+                if idx is None:
+                    idx = torch.xpu.current_device()
+                memory = int(torch.xpu.memory_allocated(idx))
+                if fraction:
+                    total = int(torch.xpu.get_device_properties(idx).total_memory)
+            except Exception:
+                pass
+            return (memory / total) if fraction and total > 0 else (memory / 2**30)
+
+        def _clear_memory(self, threshold=None):
+            if self.device.type != "xpu":
+                return _orig_clear_memory(self, threshold)
+            if threshold is not None:
+                assert 0 <= threshold <= 1, "Threshold must be between 0 and 1."
+                if self._get_memory(fraction=True) <= threshold:
+                    return
+            gc.collect()
+            if hasattr(torch.xpu, "empty_cache"):
+                torch.xpu.empty_cache()
+
+        BT._get_memory = _get_memory
+        BT._clear_memory = _clear_memory
+        BT._archery_xpu_memory_patched = True
+
+    # 3. 修补 Validator 的内存函数 (关键是添加这部分)
+    BV = ut_validator.BaseValidator
+    if not getattr(BV, "_archery_xpu_memory_patched", False):
+        # 为 Validator 添加同样的内存处理方法
+        BV._get_memory = _get_memory
+        BV._clear_memory = _clear_memory
+        BV._archery_xpu_memory_patched = True
+
+
+def _install_best_by_pixel_validate(data_yaml: str, imgsz: int, conf: float) -> None:
+    """用验证集关键点像素 mean 替代 mAP fitness，驱动 best.pt 与 patience early stopping。"""
+    import ultralytics.engine.trainer as ut
+    from ultralytics.utils import RANK
+
+    BT = ut.BaseTrainer
+    if getattr(BT, "_archery_best_by_pixel_installed", False):
+        return
+
+    _orig_validate = BT.validate
+
+    def validate(self):
+        import torch.distributed as dist
+
+        if self.ema and self.world_size > 1:
+            for buffer in self.ema.ema.buffers():
+                dist.broadcast(buffer, src=0)
+        metrics = self.validator(self)
+        if metrics is None:
+            return None, None
+        orig_fitness = metrics.pop("fitness", -self.loss.detach().cpu().numpy())
+
+        use_pixel = self.world_size <= 1 and RANK in {-1, 0}
+        mean_px: float | None = None
+        if use_pixel:
+            tmp_path: str | None = None
+            try:
+                fd, tmp_path = tempfile.mkstemp(suffix=".pt", prefix="archery_pxfit_")
+                os.close(fd)
+                from ultralytics.utils.torch_utils import unwrap_model
+
+                core = unwrap_model(self.ema.ema if self.ema else self.model)
+                torch.save({"ema": deepcopy(core).half(), "train_args": vars(self.args)}, tmp_path)
+                probe = YOLO(tmp_path)
+                stats = eval_val_pixel_error(
+                    probe,
+                    data_yaml,
+                    device=self.device,
+                    imgsz=imgsz,
+                    conf=conf,
+                )
+                mean_px = stats.get("mean_px")
+                if mean_px is None:
+                    raise RuntimeError("无有效 mean_px（检查 val 标签与检测是否为空）")
+            except Exception as exc:
+                print(f"\n⚠️ [best-by-pixel] 像素探针失败，本 epoch 仍用 mAP fitness: {exc}\n")
+                mean_px = None
+            finally:
+                if tmp_path:
+                    try:
+                        os.unlink(tmp_path)
+                    except OSError:
+                        pass
+
+        if mean_px is not None:
+            fitness = -float(mean_px)
+            metrics["metrics/mean_px(val)"] = float(mean_px)
+        else:
+            fitness = float(orig_fitness)
+
+        if not self.best_fitness or self.best_fitness < fitness:
+            self.best_fitness = fitness
+        return metrics, fitness
+
+    BT.validate = validate
+    BT._archery_best_by_pixel_installed = True
+
+
+def _fmt_csv_metric(v: float | int | None) -> str:
+    if v is None:
+        return ""
+    if isinstance(v, float):
+        return f"{v:.6g}"
+    return str(v)
+
+
+# 写入 results.csv 的列名（与 --best-by-pixel 的 metrics/mean_px(val) 区分，避免被 last.pt 回调覆盖 EMA 行）
+_PIXEL_METRIC_COLUMNS: tuple[tuple[str, str], ...] = (
+    ("pixel_error/mean_px", "mean_px"),
+    ("pixel_error/median_px", "median_px"),
+    ("pixel_error/p95_px", "p95_px"),
+    ("pixel_error/max_px", "max_px"),
+    ("pixel_error/n_points", "n_points"),
+    ("pixel_error/n_images", "n_images"),
+    ("pixel_error/skip_no_det", "skip_no_det"),
+    ("pixel_error/skip_no_gt", "skip_no_gt"),
+    ("pixel_error/skip_kpt_mismatch", "skip_kpt_mismatch"),
+)
+
+
+def _merge_pixel_metrics_into_results_csv(save_dir: str | Path, epoch_1based: int, stats: dict) -> None:
+    """在 Ultralytics 写完本 epoch 行之后，把像素指标列合并进 results.csv（扩展表头、补空列）。"""
+    csv_path = Path(save_dir) / "results.csv"
+    if not csv_path.is_file():
+        return
+    try:
+        with open(csv_path, newline="", encoding="utf-8") as f:
+            rows = list(csv.reader(f))
+    except OSError:
+        return
+    if len(rows) < 2:
+        return
+    header = list(rows[0])
+    for col_name, _ in _PIXEL_METRIC_COLUMNS:
+        if col_name not in header:
+            header.append(col_name)
+            for ri in range(1, len(rows)):
+                rows[ri].append("")
+    col_ix = {name: i for i, name in enumerate(header)}
+    rows[0] = header
+    target_ri: int | None = None
+    for ri in range(1, len(rows)):
+        row = rows[ri]
+        while len(row) < len(header):
+            row.append("")
+        try:
+            if int(float(row[0].strip())) == int(epoch_1based):
+                target_ri = ri
+        except (ValueError, IndexError):
+            continue
+    if target_ri is None:
+        return
+    row = rows[target_ri]
+    while len(row) < len(header):
+        row.append("")
+    for col_name, sk in _PIXEL_METRIC_COLUMNS:
+        row[col_ix[col_name]] = _fmt_csv_metric(stats.get(sk))
+    try:
+        with open(csv_path, "w", newline="", encoding="utf-8") as f:
+            w = csv.writer(f)
+            w.writerows(rows)
+    except OSError:
+        pass
+
+
+def _make_pixel_metrics_callback(data_yaml: str, every: int, imgsz: int, conf: float = 0.25):
+    def on_fit_epoch_end(trainer):
+        from ultralytics.utils import RANK
+
+        if RANK not in {-1, 0}:
+            return
+        if every <= 0:
+            return
+        ep = int(getattr(trainer, "epoch", -1))
+        if (ep + 1) % every != 0:
+            return
+        w = Path(trainer.save_dir) / "weights" / "last.pt"
+        if not w.is_file():
+            return
+        m = YOLO(str(w))
+        stats = eval_val_pixel_error(
+            m,
+            data_yaml,
+            device=trainer.device,
+            imgsz=imgsz,
+            conf=conf,
+        )
+        mean_px = stats.get("mean_px")
+        p95_px = stats.get("p95_px")
+        mean_s = f"{mean_px:.3f}" if mean_px is not None else "n/a"
+        p95_s = f"{p95_px:.3f}" if p95_px is not None else "n/a"
+        print(
+            f"\n[pixel-metrics] epoch {ep + 1}: mean_px={mean_s} p95_px={p95_s} "
+            f"n_points={stats.get('n_points', 0)} "
+            f"skip(det/gt/k)={stats['skip_no_det']}/{stats['skip_no_gt']}/{stats['skip_kpt_mismatch']}\n"
+        )
+        _merge_pixel_metrics_into_results_csv(trainer.save_dir, ep + 1, stats)
+
+    return on_fit_epoch_end
+
+
+def main():
+    ap = argparse.ArgumentParser(description="YOLO Pose 训练（XPU/CUDA/CPU）")
+    ap.add_argument("--data", default="datasets/dataset_pose.yaml", help="data.yaml")
+    ap.add_argument("--model", default="yolo11x-pose.pt", help="预训练权重")
+    ap.add_argument("--epochs", type=int, default=100)
+    ap.add_argument("--imgsz", type=int, default=960, help="训练输入边长（默认 960）")
+    ap.add_argument("--batch", type=int, default=4, help="批大小；OOM 时减小")
+    ap.add_argument(
+        "--device",
+        default="auto",
+        help="auto | xpu | 0 | cuda | cpu（auto：XPU 优先）",
+    )
+    ap.add_argument(
+        "--no-amp",
+        action="store_true",
+        help="关闭混合精度（默认：CUDA 开启，XPU/CPU 关闭）",
+    )
+    ap.add_argument("--project", default="runs/pose")
+    ap.add_argument("--name", default="target_pose_train")
+    ap.add_argument("--workers", type=int, default=4)
+    ap.add_argument(
+        "--pixel-metrics-every",
+        type=int,
+        default=0,
+        help="每 N 个 epoch 在 val 上打印像素误差并写入 results.csv 对应 epoch 行（0=关闭）；需 labels 与 data.yaml 布局一致",
+    )
+    ap.add_argument(
+        "--pixel-metrics-conf",
+        type=float,
+        default=0.25,
+        help="--pixel-metrics-every 时 predict 置信度阈值（默认 0.25）",
+    )
+    ap.add_argument(
+        "--best-by-pixel",
+        action="store_true",
+        help="best.pt 与 early stopping 按验证集 mean 像素误差（同 pose_pixel_metrics），fitness=-mean_px；单卡有效，DDP 自动退回 mAP",
+    )
+    ap.add_argument(
+        "--pixel-fitness-conf",
+        type=float,
+        default=0.25,
+        help="--best-by-pixel 时 predict 置信度阈值（默认与 pixel-metrics 一致）",
+    )
+    ap.add_argument(
+        "--export-onnx",
+        action="store_true",
+        help="训练结束后导出 ONNX（需再设 --onnx-imgsz）",
+    )
+    ap.add_argument(
+        "--onnx-imgsz",
+        type=int,
+        nargs=2,
+        metavar=("H", "W"),
+        default=[224, 320],
+        help="导出 ONNX 的 [高, 宽]，默认 224 320（Maix 常用）",
+    )
+    ap.add_argument(
+        "--clear-label-cache",
+        action="store_true",
+        help="启动训练前删除 data.yaml 中 path 下的 labels/*.cache（修正标注后仍报 corrupt 时用）",
+    )
+    args = ap.parse_args()
+
+    device = _pick_device(None if args.device == "auto" else args.device)
+    use_amp = False if args.no_amp else _default_amp(device)
+
+    if isinstance(device, torch.device) and device.type == "xpu":
+        print(f"✅ 使用 Intel XPU: {device}")
+    elif device == 0 or device == "0":
+        print(f"✅ 使用 CUDA: {torch.cuda.get_device_name(0)}")
+    else:
+        print("⚠️ 使用 CPU，训练会较慢")
+
+    if isinstance(device, torch.device) and device.type == "xpu":
+        _patch_ultralytics_for_xpu()
+
+    data_yaml = args.data
+    if not os.path.isabs(data_yaml):
+        data_yaml = os.path.join(os.path.dirname(os.path.abspath(__file__)), data_yaml)
+    if not os.path.exists(data_yaml):
+        print(f"❌ 数据集配置不存在: {data_yaml}")
+        return
+
+    if args.clear_label_cache:
+        n_rm = _clear_ultralytics_label_caches(data_yaml)
+        print(f"🗑️ 已删除标签目录缓存 {n_rm} 个（labels/*.cache），将强制重新扫描标注。")
+
+    print(f"📦 加载模型: {args.model}（固定 task=pose）")
+    model = YOLO(args.model, task="pose")
+
+    if args.best_by_pixel:
+        _install_best_by_pixel_validate(data_yaml, args.imgsz, args.pixel_fitness_conf)
+        print(
+            "📌 已启用 --best-by-pixel：best.pt / patience 按验证集 mean 像素误差（fitness=-mean_px）；"
+            "反向传播仍为 Ultralytics 默认 pose/box loss。"
+        )
+
+    if args.pixel_metrics_every > 0:
+        model.add_callback(
+            "on_fit_epoch_end",
+            _make_pixel_metrics_callback(
+                data_yaml, args.pixel_metrics_every, args.imgsz, conf=args.pixel_metrics_conf
+            ),
+        )
+
+    model.train(
+        task="pose",
+        data=data_yaml,
+        epochs=args.epochs,
+        imgsz=args.imgsz,
+        batch=args.batch,
+        name=args.name,
+        project=args.project,
+        exist_ok=True,
+        save=True,
+        save_period=5,
+        device=device,
+        workers=args.workers,
+        lr0=0.0001,
+        lrf=0.01,
+        optimizer="AdamW",
+        momentum=0.937,
+        weight_decay=0.001,
+        warmup_epochs=0,
+        warmup_momentum=0.8,
+        warmup_bias_lr=0.1,
+        hsv_h=0.015,
+        hsv_s=0.7,
+        hsv_v=0.4,
+        degrees=5.0,
+        translate=0.0,
+        scale=0.2,
+        shear=0.0,
+        perspective=0.0000,
+        flipud=0.0,
+        fliplr=0.5,
+        mosaic=0.0,
+        mixup=0.0,
+        copy_paste=0.0,
+        box=6,
+        cls=0.5,
+        dfl=1.5,
+        pose=18.0,
+        kobj=0.5,
+        freeze=0,
+        seed=42,
+        verbose=True,
+        amp=use_amp,
+        patience=100,
+        cos_lr=True,
+    )
+
+    print("\n✅ 训练完成！")
+    print(f"📁 best: {args.project}/{args.name}/weights/best.pt")
+    print(f"📁 last: {args.project}/{args.name}/weights/last.pt")
+    print("📊 仅看像素误差可运行: python pose_pixel_metrics.py --model <best.pt> --data <yaml> --imgsz", args.imgsz)
+
+    if args.export_onnx:
+        h, w = args.onnx_imgsz
+        print(f"📦 导出 ONNX imgsz=[{h}, {w}] ...")
+        model.export(format="onnx", imgsz=[h, w], simplify=True, opset=17, dynamic=False)
+        print("✅ ONNX 完成")
+
+
+if __name__ == "__main__":
+    main()