---
title: Detection + Segmentation Studio
emoji: 🧠
colorFrom: yellow
colorTo: red
sdk: gradio
sdk_version: 6.12.0
app_file: app.py
python_version: "3.10"
fullWidth: true
header: default
short_description: FCN, R-CNN family, Mask R-CNN, and live comparison demo.
tags:
  - gradio
  - computer-vision
  - semantic-segmentation
  - object-detection
  - instance-segmentation
  - education
---

# Detection + Segmentation Studio

An interaction-first Hugging Face Space built from the lecture `09 Detection + Segmentation`.

## What is inside

- FCN semantic segmentation with upload, overlay, class coverage, and legend
- R-CNN / Fast R-CNN / Faster R-CNN teaching studio
- Real Faster R-CNN live inference on uploaded images
- Real Mask R-CNN instance segmentation on uploaded images
- Public-metric and live-runtime comparison panel
- PDF report generation for submission

## Run locally

```powershell
python app.py
```

The default local address is `http://127.0.0.1:7860`.

## Generate the PDF report

```powershell
python tools/generate_detection_report.py
```

This creates:

- `deliverables/Detection_Segmentation_Vibe_Report.pdf`
- `deliverables/design_prompt.txt`

## Deploy to Hugging Face Spaces

Create a write-enabled token at [Hugging Face token settings](https://huggingface.co/settings/tokens), then run:

```powershell
$env:HF_TOKEN="your_token"
python deploy_hf_space.py --repo-id your-username/detection-segmentation-studio
```

The public Space URL will be:

```text
https://huggingface.co/spaces/your-username/detection-segmentation-studio
```

## Notes

- `Faster R-CNN` and `Mask R-CNN` are real pretrained inference demos.
- `R-CNN` and `Fast R-CNN` are interactive teaching simulations based on the lecture pipeline.
- The benchmark panel mixes original paper-era VOC results and current TorchVision weight-card metrics, so the app shows each method's benchmark setting explicitly.
- The app uses lazy model loading so the interface can still start before heavy models are downloaded.