yoohj0416/predictbeetle

Re-created Beetle Elytra Dataset

This dataset is a re-created version of the original 2018-NEON-beetles dataset, augmented with:

• Bounding boxes for beetle object detection.
• Manually refined elytra coordinates for precise morphological measurements.

Our goal is to streamline automated beetle trait analysis by providing curated ground truths suitable for training and benchmarking deep learning models.

What’s Included

• High-resolution images featuring multiple beetles, each with a scale bar and barcode.
• Bounding box annotations for each beetle (extracted using contour detection on Segment Anything Model (SAM) masks).
• Elytra coordinates for morphological trait estimation, specifically measuring elytra length and width.

Dataset Creation Process

1. Original Dataset

   • We started with the 2018-NEON-beetles dataset, which had images and partial morphological annotations provided via Zooniverse.

2. Segmentation & Bounding Box Extraction

   • We leveraged the Segment Anything Model (SAM) to generate masks for each beetle.
   • Contour detection was applied to masks to produce bounding box coordinates.

3. Elytra Annotation Recalibration

   • The original elytra keypoints from Zooniverse were mapped to the newly cropped bounding boxes.
   • Adjustments were made to ensure consistent coordinate references (top-left origin, pixel-based).

4. Final Verification

   • A small portion of images were manually inspected to confirm the correctness of bounding boxes and elytra points.
   • Any outliers or flawed annotations were corrected.

Intended Uses

• Object Detection: Training and evaluating models (e.g., YOLO) to detect and localize beetles in group images.
• Coordinate Regression: Predicting elytra keypoints for morphological trait analysis (length, width).
• Ecological/Biodiversity Research: Enabling large-scale measurement and comparison across beetle populations.

Citation

BibTeX:

Acknowledgements

This work was supported by the NSF OAC 2118240 Imageomics Institute award and was initiated at Beetlepalooza 2024. More details about Beetlepalooza can be found on https://github.com/Imageomics/BeetlePalooza-2024.