219 lines
6.5 KiB
TypeScript
219 lines
6.5 KiB
TypeScript
import { NextResponse } from "next/server";
|
||
import { generateText } from "ai";
|
||
import { getEnabledAiModelsWithProvider } from "@/lib/db/queries";
|
||
import { getLanguageModelById } from "@/lib/ai/dynamic-provider";
|
||
|
||
export const runtime = "nodejs";
|
||
export const maxDuration = 30;
|
||
|
||
// 位置词库
|
||
const POSITION_OPTIONS = [
|
||
'左上区域', '中央顶部', '右上区域',
|
||
'左侧中部', '中央区域', '右侧中部',
|
||
'左下区域', '中央底部', '右下区域'
|
||
] as const;
|
||
|
||
interface PickObjectRequest {
|
||
imageUrl: string; // 图片 URL
|
||
clickX: number; // 点击位置 X (0-1)
|
||
clickY: number; // 点击位置 Y (0-1)
|
||
}
|
||
|
||
interface PickObjectResult {
|
||
name: string;
|
||
position: string;
|
||
confidence: number;
|
||
boundingBox: {
|
||
x: number;
|
||
y: number;
|
||
width: number;
|
||
height: number;
|
||
};
|
||
}
|
||
|
||
export async function POST(request: Request) {
|
||
try {
|
||
const body: PickObjectRequest = await request.json();
|
||
const { imageUrl, clickX, clickY } = body;
|
||
|
||
console.log('[PickObject] 收到请求:', { imageUrl: imageUrl?.substring(0, 50), clickX, clickY });
|
||
|
||
// 验证参数
|
||
if (!imageUrl) {
|
||
return NextResponse.json({
|
||
success: false,
|
||
error: '缺少图片 URL',
|
||
}, { status: 400 });
|
||
}
|
||
|
||
if (typeof clickX !== 'number' || typeof clickY !== 'number') {
|
||
return NextResponse.json({
|
||
success: false,
|
||
error: '缺少点击坐标',
|
||
}, { status: 400 });
|
||
}
|
||
|
||
// 获取 Qwen VL 模型(复用现有逻辑)
|
||
const allModels = await getEnabledAiModelsWithProvider();
|
||
const qwenModel = allModels.find(
|
||
(m) => m.provider.type === "alibaba" && m.model.supportsVision
|
||
);
|
||
|
||
if (!qwenModel) {
|
||
console.log("[PickObject] No Qwen VL model found");
|
||
return NextResponse.json({
|
||
success: false,
|
||
error: 'Qwen VL 模型未配置',
|
||
}, { status: 500 });
|
||
}
|
||
|
||
console.log('[PickObject] 使用模型:', qwenModel.model.modelId, qwenModel.model.displayName);
|
||
|
||
const model = await getLanguageModelById(qwenModel.model.id);
|
||
if (!model) {
|
||
console.log("[PickObject] Failed to get language model");
|
||
return NextResponse.json({
|
||
success: false,
|
||
error: '无法加载模型',
|
||
}, { status: 500 });
|
||
}
|
||
|
||
// 构建 prompt(严格限制输出格式)
|
||
const prompt = `
|
||
识别图片中坐标 (${(clickX * 100).toFixed(0)}%, ${(clickY * 100).toFixed(0)}%) 位置的物品。
|
||
|
||
【输出要求 - 严格遵守】
|
||
返回纯 JSON,无其他文字:
|
||
{
|
||
"name": "物品名称",
|
||
"position": "位置描述",
|
||
"confidence": 0.95,
|
||
"boundingBox": { "x": 0.1, "y": 0.2, "width": 0.3, "height": 0.4 }
|
||
}
|
||
|
||
【name 格式】≤8字,格式:[颜色/材质]+[物品]
|
||
正确:黑色沙发、白色花瓶、木质书架、玻璃茶几
|
||
错误:一个黑色的皮质沙发(太长)
|
||
|
||
【position 格式】固定4字,从以下选择:
|
||
${POSITION_OPTIONS.join('、')}
|
||
|
||
【boundingBox】相对坐标 0-1,精确到小数点后2位
|
||
`.trim();
|
||
|
||
console.log('[PickObject] 调用 Qwen VL 模型...');
|
||
|
||
// 调用 Qwen VL 模型(复用现有的 generateText)
|
||
const { text: responseText } = await generateText({
|
||
model,
|
||
messages: [
|
||
{
|
||
role: 'user',
|
||
content: [
|
||
{
|
||
type: 'image',
|
||
image: imageUrl, // 直接使用图片 URL
|
||
},
|
||
{
|
||
type: 'text',
|
||
text: prompt,
|
||
},
|
||
],
|
||
},
|
||
],
|
||
temperature: 0.1,
|
||
maxTokens: 300,
|
||
} as any);
|
||
|
||
console.log('[PickObject] Qwen 返回内容:', responseText?.substring(0, 200));
|
||
|
||
if (!responseText) {
|
||
return NextResponse.json({
|
||
success: false,
|
||
error: 'AI 未返回结果',
|
||
}, { status: 500 });
|
||
}
|
||
|
||
// 解析 JSON 结果
|
||
let result: PickObjectResult;
|
||
try {
|
||
// 尝试从返回内容中提取 JSON
|
||
const jsonMatch = responseText.match(/\{[\s\S]*\}/);
|
||
if (!jsonMatch) {
|
||
throw new Error('未找到 JSON');
|
||
}
|
||
result = JSON.parse(jsonMatch[0]);
|
||
|
||
// 验证和修正结果
|
||
if (!result.name || result.name.length > 12) {
|
||
result.name = result.name?.substring(0, 8) || '未知物品';
|
||
}
|
||
|
||
// 验证位置是否在词库中
|
||
if (!POSITION_OPTIONS.includes(result.position as any)) {
|
||
// 根据点击位置推算
|
||
const posX = clickX < 0.33 ? '左' : clickX > 0.66 ? '右' : '中央';
|
||
const posY = clickY < 0.33 ? '上' : clickY > 0.66 ? '下' : '中';
|
||
if (posX === '中央') {
|
||
result.position = posY === '中' ? '中央区域' : posY === '上' ? '中央顶部' : '中央底部';
|
||
} else {
|
||
result.position = posY === '中' ? `${posX}侧中部` : `${posX}${posY}区域`;
|
||
}
|
||
}
|
||
|
||
// 确保 boundingBox 有效
|
||
if (!result.boundingBox ||
|
||
typeof result.boundingBox.x !== 'number' ||
|
||
typeof result.boundingBox.y !== 'number') {
|
||
result.boundingBox = {
|
||
x: Math.max(0, clickX - 0.15),
|
||
y: Math.max(0, clickY - 0.15),
|
||
width: 0.3,
|
||
height: 0.3,
|
||
};
|
||
}
|
||
|
||
// 确保 confidence 有效
|
||
if (typeof result.confidence !== 'number' || result.confidence < 0 || result.confidence > 1) {
|
||
result.confidence = 0.85;
|
||
}
|
||
|
||
} catch (parseError) {
|
||
console.error('[PickObject] JSON parse error:', parseError, 'Content:', responseText);
|
||
|
||
// 返回默认结果
|
||
const posX = clickX < 0.33 ? '左' : clickX > 0.66 ? '右' : '中央';
|
||
const posY = clickY < 0.33 ? '上' : clickY > 0.66 ? '下' : '中';
|
||
const position = posX === '中央'
|
||
? (posY === '中' ? '中央区域' : posY === '上' ? '中央顶部' : '中央底部')
|
||
: (posY === '中' ? `${posX}侧中部` : `${posX}${posY}区域`);
|
||
|
||
result = {
|
||
name: '物品',
|
||
position,
|
||
confidence: 0.5,
|
||
boundingBox: {
|
||
x: Math.max(0, clickX - 0.15),
|
||
y: Math.max(0, clickY - 0.15),
|
||
width: 0.3,
|
||
height: 0.3,
|
||
},
|
||
};
|
||
}
|
||
|
||
console.log('[PickObject] 识别结果:', result);
|
||
|
||
return NextResponse.json({
|
||
success: true,
|
||
object: result,
|
||
});
|
||
|
||
} catch (error) {
|
||
console.error('[PickObject] Error:', error);
|
||
return NextResponse.json({
|
||
success: false,
|
||
error: `识别失败: ${error instanceof Error ? error.message : '未知错误'}`,
|
||
}, { status: 500 });
|
||
}
|
||
}
|