Skip to content

chore: Update README #34

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
dronefreak merged 1 commit into from
Jul 13, 2026
Merged

chore: Update README #34

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 11 additions & 29 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,24 +14,6 @@ CABiNet (Context Aggregation Network) is a dual-branch convolutional neural netw
<img src="assets/showcase_mosaic.gif" alt="UAVid Semantic Segmentation Demo">
</p>

## Key Features

- **High Performance**: Achieves 75.9% mIoU on Cityscapes test set at 76 FPS (NVIDIA RTX 2080Ti)
- **Edge Deployment**: 8 FPS on Jetson Xavier NX for embedded applications
- **Dual-Branch Architecture**: Combines high-resolution spatial detailing with efficient context aggregation
- **Lightweight Design**: Reduced computational overhead through optimized global and local context blocks
- **Multi-Scale Support**: Effective feature extraction across different scales

## Performance

| Model | Dataset | mIoU | FPS (RTX 2080Ti) | FPS (Jetson Xavier NX) | Params (M) |
| ------------------ | ---------- | ----- | ---------------- | ---------------------- | ---------- |
| **CABiNet (ours)** | Cityscapes | 75.9% | 76 | 8 | ~3.0 |
| YOLOv8m-seg | Cityscapes | 73.5% | — | — | 27.3 |
| YOLOv11m-seg | Cityscapes | 74.2% | — | — | 22.4 |

Cityscapes numbers above are literature-reported instance-segmentation (`-seg`) baselines, kept for historical reference. For UAVid, CABiNet is benchmarked directly against Ultralytics' dense semantic-segmentation task (`yolo26*-sem`, not `-seg`) under one shared pipeline — see [UAVid Model Zoo](#uavid-model-zoo) for the full per-model comparison.

## Architecture

The CABiNet architecture employs a dual-branch design that balances spatial detail preservation and contextual understanding:
Expand Down Expand Up @@ -63,17 +45,17 @@ _Columns: Input images, State-of-the-art predictions, CABiNet predictions (white

## UAVid Model Zoo

CABiNet and Ultralytics YOLO26 (`semantic` task, dense per-pixel — not `-seg`) are trained and evaluated under one shared UAVid pipeline (`images/`+`masks/` format, 8 classes incl. Clutter). Full per-class breakdowns and model cards: [`hf_modelcards/`](hf_modelcards/); HF weights below will be populated once the models are pushed.

| Model | mIoU (%) | Params (M) | FLOPs (B) | HF Weights |
| --------------------------- | -------- | ---------- | --------- | ---------- |
| CABiNet (MobileNetV3-Large) | TBD¹ | TBD | TBD | TBD |
| CABiNet (MobileNetV3-Small) | TBD² | TBD | TBD | TBD |
| YOLO26n-sem | 63.58 | TBD | TBD | TBD |
| YOLO26s-sem | 66.88 | TBD | TBD | TBD |
| YOLO26m-sem | 67.66 | TBD | TBD | TBD |
| YOLO26l-sem | 67.20 | TBD | TBD | TBD |
| YOLO26x-sem | TBD³ | TBD | TBD | TBD |
CABiNet and Ultralytics YOLO26 (`semantic` task, dense per-pixel — not `-seg`) are trained and evaluated under one shared UAVid pipeline (`images/`+`masks/` format, 8 classes incl. Clutter).

| Model | mIoU (%) | Params (M) | FLOPs (B) | HF Weights |
| --------------------------- | -------- | ---------- | --------- | ----------------------------------------------------------------------------- |
| CABiNet (MobileNetV3-Large) | TBD¹ | TBD | TBD | [HF Model](https://huggingface.co/dronefreak/cabinet-mobilenetv3-small-uavid) |
| CABiNet (MobileNetV3-Small) | TBD² | TBD | TBD | [HF Model](https://huggingface.co/dronefreak/cabinet-mobilenetv3-large-uavid) |
| YOLO26n-sem | 63.58 | TBD | TBD | [HF Model](https://huggingface.co/dronefreak/uavid-yolo26n-sem) |
| YOLO26s-sem | 66.88 | TBD | TBD | [HF Model](https://huggingface.co/dronefreak/uavid-yolo26s-sem) |
| YOLO26m-sem | 67.66 | TBD | TBD | [HF Model](https://huggingface.co/dronefreak/uavid-yolo26m-sem) |
| YOLO26l-sem | 67.20 | TBD | TBD | [HF Model](https://huggingface.co/dronefreak/uavid-yolo26l-sem) |
| YOLO26x-sem | TBD³ | TBD | TBD | [HF Model](https://huggingface.co/dronefreak/uavid-yolo26x-sem) |

¹ Trained under the refactored pipeline but not yet re-evaluated. ² Not yet trained. ³ Not yet trained.

Expand Down
Loading