SlopeSense

SlopeSense is a computer-vision pipeline developed for the KITE/IDAPT research group at UHN. It classifies whether participants pass or fail a walking-on-slopes trial using GoPro video footage and skeleton pose sequences fed through a graph neural network (CTR-GCN).

The project ships as two separate tools:

Tool	Purpose
SlopeSense Renaming App	Matches raw GoPro videos to MATLAB tipper files, applies human-in-the-loop corrections, renames outputs, and runs CV inference to validate pass/fail classification
SlopeSense Training Tool	Full ML training pipeline: pose extraction → dataset building → hyperparameter search → model training

---

Downloads

Pre-built Windows executables are available on the Releases page.

Asset	Size	Notes
SlopeSense-GPU.part1	~1.9 GB	Training tool — GPU build, part 1 of 2
SlopeSense-GPU.part2	~1.4 GB	Training tool — GPU build, part 2 of 2
assemble_SlopeSense_GPU.bat	<1 KB	Reassembly script for the GPU build parts
SlopeSense-CPU.part1	~1.9 GB	Training tool — CPU only build, part 1 of 2
SlopeSense-CPU.part2	~1.4 GB	Training tool — CPU only build, part 2 of 2
assemble_SlopeSense_CPU.bat	<1 KB	Reassembly script for the CPU build parts
CTR-GCN.zip	~9.5 MB	External CTR-GCN framework (required by training tool)
yolo26x-pose.pt	~121 MB	YOLO pose model weights

An NVIDIA GPU is strongly recommended for the GPU build. For older cards (e.g. GTX 1080 Ti), the included CUDA 11.8 build is suitable.

---

Setting Up the Training Tool (SlopeSense GPU Build)

Step 1 — Download the parts

Download all three files from the Releases page into the same folder:

SlopeSense-GPU.part1
SlopeSense-GPU.part2
assemble_SlopeSense_GPU.bat

Step 2 — Reassemble the ZIP

Double-click assemble_SlopeSense_GPU.bat. It will join the two parts back into SlopeSense-GPU.zip automatically. A window will appear — wait for the "Done!" message.

Step 3 — Extract

Right-click SlopeSense-GPU.zip → Extract All → choose a destination (e.g. C:\SlopeSense\).

Step 4 — Add the external dependencies

Also download and place alongside the app:

• CTR-GCN.zip → extract to frameworks\CTR-GCN\
• yolo26x-pose.pt → place anywhere accessible (you will point the app to it in Config)

A reasonable folder layout:

C:\SlopeSense\
  SlopeSense-GPU\        ← extracted app folder
  frameworks\
    CTR-GCN\             ← extracted CTR-GCN repo
  models\
    yolo26x-pose.pt
  projects\
    my_project\
      videos\
      outputs\
      runs\
      production\

Step 5 — Launch and configure

1. Open SlopeSense-GPU\SlopeSense.exe
2. In the Config tab, set:
   • Video Folder
   • Pose Output Folder
   • Runs Output Folder
   • Production Output Folder
   • YOLO Model → point to yolo26x-pose.pt
   • CTR-GCN Repo → point to frameworks\CTR-GCN\
3. Click Save Project
4. Work left to right: Videos → Poses → Training → Production

For the full training workflow reference, see kite-computer-vision-slip-detector/README.md.

---

Setting Up the Training Tool (SlopeSense CPU Build)

The CPU build setup is identical to the GPU build above, substituting CPU for GPU everywhere:

1. Download SlopeSense-CPU.part1, SlopeSense-CPU.part2, and assemble_SlopeSense_CPU.bat into the same folder
2. Double-click assemble_SlopeSense_CPU.bat — wait for the "Done!" message
3. Right-click SlopeSense-CPU.zip → Extract All
4. Follow Steps 4–5 from the GPU setup above

The CPU build runs without an NVIDIA GPU but pose extraction and training will be significantly slower.

---

Setting Up the Renaming App

The Renaming App is built separately and distributed as its own package. See the Client User Guide for complete setup and usage instructions.

The guide covers:

• Expected folder structure for videos and tipper files
• Running the renaming workflow
• Running CV validation
• Handling HITL dialogs
• Reading the output reports

---

Building From Source

Both tools include build scripts for developers.

Training Tool (kite-computer-vision-slip-detector)

# From repo root
cd kite-computer-vision-slip-detector
python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt
python -m gui

Or to produce the .exe:

.\build_windows.ps1

See kite-computer-vision-slip-detector/README.md for GPU-enabled build instructions.

Renaming App

# From repo root
.\build_windows.ps1

Produces Renaming Application\dist\SlopeSense\SlopeSense.exe.

---

Repository Structure

aps490-capstone-kite/
├── Renaming Application/          # Operational tool: video matching + CV validation
│   └── RenamingApp/
│       └── CLIENT_USER_GUIDE.md   # End-user guide for the Renaming App
├── kite-computer-vision-slip-detector/  # ML training pipeline
│   ├── README.md                  # Training tool documentation
│   ├── dist/                      # Pre-built executables (gitignored)
│   └── frameworks/CTR-GCN/        # External CTR-GCN graph network
├── models/                        # Shared model weights (gitignored)
├── build_windows.ps1              # Windows build script (Renaming App)
└── build_macos.sh                 # macOS build script (Renaming App)

---

System Requirements

• Windows 10 or Windows 11
• NVIDIA GPU recommended (required for GPU build)
• No Python installation needed when using pre-built executables
• ~4 GB free disk space for the GPU build after extraction