https://github.com/shenliuming/inlook-yolo-model-lab
INLOOK Studio is an open-source AI content creation workbench for turning owned or authorized scripts, audio, and video materials into reusable talking-head content.
https://github.com/shenliuming/inlook-yolo-model-lab
ai-education computer-vision model-testing object-detection ultralytics yolo yolov8
Last synced: 9 days ago
JSON representation
INLOOK Studio is an open-source AI content creation workbench for turning owned or authorized scripts, audio, and video materials into reusable talking-head content.
- Host: GitHub
- URL: https://github.com/shenliuming/inlook-yolo-model-lab
- Owner: shenliuming
- Created: 2026-05-31T06:42:52.000Z (about 1 month ago)
- Default Branch: main
- Last Pushed: 2026-06-18T10:03:30.000Z (19 days ago)
- Last Synced: 2026-06-18T11:29:01.000Z (19 days ago)
- Topics: ai-education, computer-vision, model-testing, object-detection, ultralytics, yolo, yolov8
- Language: Python
- Homepage: https://in-look.cn/yolo/
- Size: 6.59 MB
- Stars: 2
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# INLOOK YOLO Model Lab
一个用于图片、视频和摄像头识别测试的本地 YOLO 网页实验平台。
INLOOK YOLO Model Lab is a local-first YOLO web lab for image, video, and camera-based recognition tests. It is designed for computer vision learning, model validation, demo recording, and lightweight internal tooling.
## Online Demo
- [https://in-look.cn/yolo/](https://in-look.cn/yolo/)
## Architecture Docs
- [正式架构说明](docs/architecture.md)
- [项目架构图 + 数据流图 + 目录职责说明](docs/project_architecture_and_data_flow.md)
## Frontend Roles
- `inlook-studio-web` 是正式产品主前端
- `apps/yolo-web` 保留为实验室前端
## Features
- 图片识别
- 视频识别
- 摄像头识别
- OBS 虚拟摄像头
- 模型切换
- 运行日志
- JSON 测试报告
- 结果下载
- 字幕识别
- TTS 配音生成
## Screenshots
Project screenshots can be placed here:
- `docs/images/home.png`
- `docs/images/yolo-demo.png`
Current repository state:
- Screenshot placeholders only
- No fake images generated in README
## Quick Start
### 1. Install `uv`
```bash
brew install uv
```
### 2. Install backend dependencies
```bash
cd apps/yolo-api
uv venv --python 3.11
source .venv/bin/activate
uv pip install -r requirements.txt
```
### 2.1 Prepare CosyVoice for local TTS
```bash
cd apps/yolo-api
uv pip install -r requirements.txt
```
Configure the CosyVoice model directory before starting the backend:
```bash
export TTS_ENGINE=cosyvoice
export COSYVOICE_MODEL_DIR=pretrained_models/CosyVoice2-0.5B
export COSYVOICE_DEVICE=auto
export COSYVOICE_SAMPLE_RATE=24000
```
INLOOK Studio now routes builtin voices, custom voices, and current-video voices
through CosyVoice only. MOSS-TTS is deprecated and is not used as a fallback.
### 3. Start FastAPI
```bash
cd apps/yolo-api
uv run uvicorn app:app --reload --host 127.0.0.1 --port 7860
```
### 4. Start frontend
```bash
cd apps/yolo-web
npm install
npm run dev -- --host 127.0.0.1 --port 5173
```
### 5. Open the local page
- Frontend: `http://127.0.0.1:5173`
- Backend health: `http://127.0.0.1:7860/api/health`
If you want to use the old local backend port `8000`, start Vite with:
```bash
VITE_API_TARGET=http://127.0.0.1:8000 npm run dev -- --host 127.0.0.1 --port 5173
```
## Compliance
本项目仅用于计算机视觉学习、模型测试和内容创作。
系统只输出识别结果，不提供任何游戏控制、自动操作或绕过机制。
## Want To Train Your Own Model?
如果你正在训练自己的 YOLO 模型，可以先准备这些材料：
- `data.yaml`
- 训练结果图
- 少量样本图
- 测试视频
这样更容易判断问题可能出在：
- 数据本身
- 标注质量
- 类别设计
- 训练参数
- 场景差异
## Project Structure
```text
inlook-yolo-model-lab/
├── apps/
│ ├── yolo-api/
│ │ ├── app.py
│ │ ├── app/
│ │ │ ├── main.py
│ │ │ ├── controllers/
│ │ │ ├── services/
│ │ │ ├── clients/
│ │ │ ├── config/
│ │ │ ├── common/
│ │ │ └── utils/
│ │ ├── models/
│ │ ├── requirements.txt
│ │ └── Dockerfile
│ └── yolo-web/
│ ├── src/api/
│ ├── src/components/
│ ├── src/App.vue
│ └── Vue frontend
├── assets/
│ └── demo/
├── docs/
├── deploy/
├── pretrained_models/
│ └── CosyVoice2-0.5B/ # local model directory, not committed
└── README.md
```
## AI Content Workflow
The content workflow is separate from the YOLO vision lab. It currently includes:
- Material Intake
- Subtitle Recognition
- TTS Voice Generation
Subtitle-related files are integrated into the backend service layer:
- `apps/yolo-api/app/services/subtitle_tool/subtitle_pack.py`
- `apps/yolo-api/app/services/subtitle_tool/burn_subtitles.py`
- `apps/yolo-api/app/services/subtitle_tool/check_env.py`
TTS-related files:
- `apps/yolo-api/app/controllers/tts_controller.py`
- `apps/yolo-api/app/services/tts_service.py`
- `apps/yolo-api/app/services/tts_engines/cosyvoice_engine.py`
TTS runtime output:
- `apps/yolo-api/runtime/content_lab/tts/tasks/{task_id}/inputs`
- `apps/yolo-api/runtime/content_lab/tts/tasks/{task_id}/outputs`
- `apps/yolo-api/runtime/content_lab/tts/tasks/{task_id}/run.log`
Helpful docs:
- `docs/subtitle-workflow/example_usage.md`
- `docs/subtitle-workflow/PRODUCTION_WORKFLOW.md`
Quick env check:
```bash
uv run python apps/yolo-api/app/services/subtitle_tool/check_env.py
```
## Backend
Recommended startup:
```bash
cd apps/yolo-api
uv venv --python 3.11
uv pip install -r requirements.txt
uv run uvicorn app:app --reload --host 127.0.0.1 --port 7860
```
API endpoints:
- `GET /api/v1/health`
- `GET /api/v1/vision/health`
- `GET /api/v1/vision/models`
- `POST /api/v1/vision/models/select`
- `POST /api/v1/vision/images/detect`
- `POST /api/v1/vision/videos/detect`
- `POST /api/v1/vision/realtime/detect`
- `GET /api/v1/vision/tasks/{task_id}`
- `GET /api/v1/vision/tasks/{task_id}/files/{filename}`
- `GET /api/v1/content-lab/health`
- `GET /api/v1/content-lab/materials/health`
- `POST /api/v1/content-lab/materials/tasks`
- `GET /api/v1/content-lab/materials/tasks/{task_id}`
- `GET /api/v1/content-lab/materials/tasks/{task_id}/files/{filename}`
- `GET /api/v1/content-lab/subtitles/health`
- `POST /api/v1/content-lab/subtitles/tasks`
- `GET /api/v1/content-lab/subtitles/tasks/{task_id}`
- `POST /api/v1/content-lab/subtitles/tasks/{task_id}/reburn`
- `GET /api/v1/content-lab/subtitles/tasks/{task_id}/files/{filename}`
- `GET /api/v1/content-lab/tts/health`
- `POST /api/v1/content-lab/tts/tasks`
- `GET /api/v1/content-lab/tts/tasks/{task_id}`
- `GET /api/v1/content-lab/tts/tasks/{task_id}/files/{filename}`
Compatibility endpoints are still available:
- `GET /api/models`
- `POST /api/detect/image`
- `POST /api/detect/video`
- `POST /api/realtime/detect`
- `POST /api/materials/tasks`
- `GET /api/materials/tasks/{task_id}`
- `GET /api/materials/tasks/{task_id}/files/{filename}`
Static result paths:
- `/outputs`
- `/reports`
## Frontend
```bash
cd apps/yolo-web
npm install
npm run dev -- --host 127.0.0.1 --port 5173
```
Open:
- `http://127.0.0.1:5173`
Frontend routes:
- `/`
- `/vision-lab`
- `/vision-lab/model-test`
- `/vision-lab/image`
- `/vision-lab/video`
- `/vision-lab/realtime`
- `/content-workflow`
- `/content-workflow/material-intake`
- `/content-workflow/subtitle-recognition`
- `/content-workflow/tts`
- `/content-lab`
- `/content-lab/material-intake`
- `/content-lab/subtitle-recognition`
- `/content-lab/tts`
Legacy redirects:
- `/material-intake` -> `/content-workflow/material-intake`
- `/content-intake` -> `/content-workflow/material-intake`
- `/vision-lab/image-detect` -> `/vision-lab/image`
- `/vision-lab/video-detect` -> `/vision-lab/video`
- `/vision-lab/realtime-detect` -> `/vision-lab/realtime`
Shared frontend modules:
- `src/api/client.js`
- `src/api/vision.js`
- `src/api/workflow.js`
- `src/api/contentLabApi.js`
- `src/components/StatusCard.vue`
- `src/components/TaskLog.vue`
- `src/components/FileDownloadList.vue`
## Camera Mode
- Frontend uses `getUserMedia()` to open camera devices
- Supports standard cameras and OBS virtual camera selection
- Captures frames and sends them to the backend for YOLO inference
- Draws boxes, labels, and confidence values on overlay canvas
- Shows the latest JSON result in realtime mode
## Models
Default custom model:
- `apps/yolo-api/models/inlook/best.pt`
Scanned model directories:
- `apps/yolo-api/models/official/*.pt`
- `apps/yolo-api/models/inlook/*.pt`
Security-related constraints:
- Model files stay in backend-only storage
- Nginx does not expose model directories
- FastAPI does not mount model directories as static paths
- `/api/models` returns display metadata instead of real model file paths
- Report JSON does not expose real model file paths
- Image upload limit: `10MB`
- Video upload limit: `200MB`
- Realtime frame upload limit: `4MB`
- Allowed image types: `jpg/jpeg/png`
- Allowed video type: `mp4`
- Basic IP rate limiting is enabled in backend
- Old uploads / outputs / reports are cleaned periodically
- Optional `INLOOK_API_KEY` is supported for internal access control
## Internal API Key
If you want to restrict who can call the backend:
```bash
INLOOK_API_KEY=your-secret-key
```
If frontend should attach the same key automatically before build:
```bash
VITE_INTERNAL_API_KEY=your-secret-key
```
The frontend will send:
```txt
X-INLOOK-Key: your-secret-key
```
## Docker Deployment
The project already includes a Docker-based deployment structure for ECS / VM deployment.
Files already included:
- `apps/yolo-api/Dockerfile`
- `apps/yolo-web/Dockerfile`
- `deploy/nginx.conf`
- `docker-compose.yml`
- `.dockerignore`
### Before deployment
Make sure the server already has:
- Docker
- Docker Compose Plugin
- Model file `apps/yolo-api/models/inlook/best.pt`
Optional official models:
- `apps/yolo-api/models/official/yolo11n.pt`
- `apps/yolo-api/models/official/yolo11s.pt`
- `apps/yolo-api/models/official/yolov8n.pt`
### Start
```bash
docker compose up -d --build
```
After startup:
- Frontend: `http://your-server-ip/`
- Backend health: `http://your-server-ip/api/health`
### Stop
```bash
docker compose down
```
### Docker cache note
For normal backend updates:
```bash
docker compose build backend
docker compose up -d backend
```
Avoid using:
```bash
docker compose build --no-cache backend
```
because it forces large dependencies such as `torch`, `ultralytics`, and `opencv` to be downloaded again.
## Deploy Under `in-look.cn/yolo/`
If you want to mount the frontend under an existing website path:
- Page entry: `https://in-look.cn/yolo/`
- Backend API: `https://in-look.cn/yolo/api/*`
- Result files: `https://in-look.cn/yolo/outputs/*`
- Report files: `https://in-look.cn/yolo/reports/*`
Build frontend with subpath base:
```bash
cd apps/yolo-web
VITE_PUBLIC_BASE=/yolo/ npm run build
```
Then publish `apps/yolo-web/dist/` to:
```bash
/var/www/in-look.cn/html/yolo/
```
## Notes
Please do not commit:
- model files
- videos
- audio files
- training datasets
- generated subtitle outputs
- generated recognition outputs
## INLOOK Studio LLM 配置
提示词改写、文案校对和标题生成统一走 OpenAI-compatible Chat Completions 服务。不要把真实 API Key 写进代码，使用环境变量：
```bash
LLM_PROVIDER=openai_compatible
LLM_BASE_URL=https://your-model-service/v1
LLM_API_KEY=your_api_key
LLM_MODEL=your-model-name
LLM_TIMEOUT_SECONDS=60
```
未配置时 `GET /api/v1/ai/status` 会返回 `available=false`，前端会禁用 AI 改写按钮，不会生成 mock 文案。