BurrowScope - Bank Swallow Burrow Segmentation

Quick start

Visual inspection workflow

1 Choose one bank photo, selected examples, or a local image set to inspect.
2 Run the ONNX model in your browser and keep every image local.
3 Review post-NMS boxes, masks, centroids, and confidence behaviour.
4 Export CSV, JSON, or an overlay PNG for local field records.

Image input

Bank photo

Place your own bank-wall images here to analyse them with the burrow model.

Upload Bank Photo JPG, PNG, or WebP. You can select several with Ctrl/Cmd.

No image loaded

Choose a bank-wall image first.

Model settings

Detection engine

Local ONNX

Model file

bank_swallow_burrow_yolo11s_seg_production.onnx

Waiting for ONNX Runtime

If there is an updated inference model, upload it here.

Confidence Threshold 0.35 IoU NMS Threshold 0.45 Mask Threshold 0.55 Maximum detections

Example images to learn how it works

Try sample bank photos

Batch results

0 burrows

Image	Burrows

Visual workspace

Per-image burrow detection

No analysed images in this session.

Masks, boxes, and centroids render here.

Detections dashboard

Burrows Detected

0 post-NMS boxes

Mask status

Counts report decoded post-NMS boxes; mask availability can be reported separately.

ID	Confidence	Centroid	Bounding box	Mask px / status
No predictions yet.

Threshold diagnostics

Decode, NMS, and mask checks are tracked independently.

Confidence cutoff	0.35
IoU NMS cutoff	0.45
Mask cutoff	0.55
Max detections	500
Decoded boxes	Awaiting inference
Post-NMS boxes	0
Mask availability	Awaiting inference

Bank Swallow burrow inspection

Per-image burrow detection

Burrows Detected

Session

Use conditions

Select example images

BurrowScope glossary

YOLO

YOLO-seg

Transfer learning

ONNX

ONNX Runtime Web

NMS

IoU

Confidence threshold

Mask threshold

Bounding box and centroid