fix: YOLO model path, Docker CPU optimization, remove dead import

[kayceepeece]

Context

Deploying on a VPS (Docker) hits 3 bugs that prevent the app from running.

Changes

1. Make YOLO model path configurable (main.py + Dockerfile)

• YOLO("yolov8n.pt") tries to download to CWD, but /app is volume-mounted in docker-compose, so the app user has no write permission → PermissionError
• Add YOLO_MODEL_PATH env var (default: yolov8n.pt for bare-metal auto-download)
• Docker sets /tmp/Ultralytics/yolov8n.pt (pre-downloaded during build, survives volume mount)
• Bare-metal users get auto-download to CWD — no breakage

2. CPU-only PyTorch by default in Docker (Dockerfile)

• Split pip install: torch via --index-url (explicit), rest via requirements.txt
• TORCH_INDEX_URL build arg (default: https://download.pytorch.org/whl/cpu)
• GPU users: docker compose build --build-arg TORCH_INDEX_URL=https://download.pytorch.org/whl/cu124
• Image: 11.3 GB → 4.6 GB

This follows the same pattern as PyTorch own Dockerfile which uses ARG CUDA_PATH=cu121.

3. Remove unused import scenedetect (main.py)

Dead code after the open_video() refactor — no scenedetect. references remain.

Testing

• docker compose build succeeds (4.6 GB image)
• Backend starts and serves API on :8000
• YOLO model loads from env-configured path
• requirements.txt unchanged — bare-metal pip install still works

[Copilot]

Pull request overview

This PR addresses Docker/VPS deployment blockers by making the Ultralytics YOLO model location configurable, optimizing Docker builds for CPU-only PyTorch by default, and removing a dead import.

Changes:

• Added YOLO_MODEL_PATH env support in main.py to control where the YOLO model is loaded from.
• Updated Dockerfile to install CPU-only PyTorch via a configurable index URL and to pre-download the YOLO model into /tmp/Ultralytics.
• Removed an unused import scenedetect from main.py.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
main.py	Reads YOLO_MODEL_PATH from environment when instantiating the YOLO model; removes a redundant scenedetect import.
Dockerfile	Adds CPU-only PyTorch installation flow and pre-downloads the YOLO model during build for Docker deployments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.