Related Publication: Detection of Strong Gravitational Lenses with Lens-Topology Relation and Arc-Assisted Annotation
Data & Annotation Strategy:
Sourced from DESI DR10. Unlike traditional datasets annotating only gravitational lens systems, we employ a gravitational lens arc-assisted strategy. We provide bounding boxes for both the entire gravitational lens system and the auxiliary gravitational lens arc, leveraging arc morphology to improve accuracy and recall.
File Mapping:
1. images.zip: 16,440 Numpy arrays (.npy, 256x256, 4 bands: g,r,i,z). Includes 6,440 positive (1,288 lenses x 5 jitters) and 10,000 negative samples (2,000 non-lenses, including LRGs/ring galaxies).
2. instances_train.json: COCO annotations for 11,505 training images.
3. instances_val.json: COCO annotations for 1,640 validation images.
4. instances_test.json: COCO annotations for 3,295 test images.

(Note: See README.txt for detailed dimensions and naming conventions.)