ObjectDetect4Blind

A multi-threaded pipeline that runs three vision models in parallel — Depth Estimation (Depth-Anything-V2), Object Detection (YOLO), and Segmentation — on images or video. Outputs are saved to ./output.

---

Features

• Orchestrates depth, detection, and segmentation together via MAIN.py
• Supports image and video inputs
• Simple, folder-based model checkpoint placement

---

Project Layout

RECOMMEND: put file respitory the same as me, or u need to fix code for file/model location

C:/Python/
├────ObjectDetect4Blind/
│    ├── .vscode
│    ├── MAIN.py                         # Launches multithreading 3 models
│    ├── assets/                         # Example input images
│    ├── output/                         # Outputs produced by MAIN.py
│    ├── Depth-Anything-V2-main/         # Depth estimation module
│    │   ├── app.py
│    │   ├── run.py
│    │   └──run_video.py                 # Temp not developed
│    ├── Object detection/               # Object detection module
│    │   └── main.py                     # Usage entrypoint for detection
│    └── Segmentation/                   # Segmentation module
│        └── test_model.py               # Usage entrypoint for segmentation
└────ObjectDetectRequireFile/
     ├── put-in-depth-anything
     ├── put-in-obj-detect                   
     ├── put-in-segment                   
     └── output/ 

Setup

Model file

https://drive.google.com/file/d/1DwhseV8bqV_qw7CIuWS7pnMRJ9au9lpE/view

Python versions (important)

This repo currently expects two Python interpreters when running MAIN.py:

• YOLO / Object detection: Python 3.11
• Depth estimation: Python 3.13

MAIN.py starts each model using the interpreter paths you configure.
Edit MAIN.py and replace the hard-coded interpreter/virtual-env paths with the ones on your machine (see comments in the file).

✅ Tip: If you manage multiple Python versions, set up two virtual environments (e.g., with Conda or pyenv) and point MAIN.py to their python executables.

---

Quickstart

A) Run the full pipeline (multithreading)

1. Open MAIN.py and update the Python paths/env activation commands for:
   • Detection, Segmentation (Python 3.11)
   • Depth (Python 3.13)
2. From the project root: python MAIN.py

B) Run Depth-Anything-V2 only

1. Local demo server
   • python app.py
2. Single img(depth estimation only)
   • python run.py --encoder vits --precision int8 --img-path "C:\Python\ObjectDetect4Blind\assets\demo01.jpg" --outdir depth_vis --pred-only
3. Single image (side-by-side input + depth)
   • python run.py --encoder vits --precision int8 --img-path "C:\Python\ObjectDetect4Blind\assets\demo01.jpg" --outdir depth_vis
4. Video
   • python run_video.py --encoder vitl \ --video-path assets/examples_video \ --outdir video_depth_vis