Images40
Instances122
Classes3
Tasksegment
Background imgs2
Issues (auto-fixed)6 (6)
Recommendations
- Small dataset: start from pretrained weights (e.g. yolo26n.pt / yolo26s.pt), keep default augmentations, and expect noisy val metrics - use the test split once at the end.
- Segmentation dataset detected: train a *-seg model (e.g. yolo26n-seg.pt).
Class distribution by split
Split balance
| Class | train | val | test | Total |
|---|---|---|---|---|
| 0 circle | 58 (82%) | 5 (7%) | 8 (11%) | 71 |
| 1 square | 25 (74%) | 4 (12%) | 5 (15%) | 34 |
| 2 triangle | 13 (76%) | 2 (12%) | 2 (12%) | 17 |
| Images | 30 imgs | 5 imgs | 5 imgs | 40 |
✓ No group leakage between splits.
Geometry
Box size (relative)
Box centre heatmap
Aspect ratio (w/h)Objects per image
Size buckets at imgsz=640 (COCO definition): small 0 · medium 54 · large 68. Median box area = 2.75% of the image.
| Image size | Count |
|---|---|
| 320×240 | 40 |
Validation
| Severity | Check | Count | Auto-fixed |
|---|---|---|---|
| warning | duplicate_box | 1 | 1 |
| warning | duplicate_image | 1 | 1 |
| warning | mixed_box_seg | 1 | 1 |
| warning | out_of_bounds | 1 | 1 |
| warning | tiny_box | 1 | 1 |
| info | bbox_polygon_mismatch | 1 | 1 |
Show details (6 of 6)
| Sev | Check | File | Message | Fixed |
|---|---|---|---|---|
| warning | duplicate_box | subj02_frame0000.jpg | obj#4: duplicates another 'circle' (IoU>0.9); dropped | ✓ |
| warning | duplicate_image | subj99_copy.jpg | identical bytes to subj05_frame0000.jpg (train/val leakage risk); dropped | ✓ |
| warning | mixed_box_seg | subj04_frame0000.jpg | 1 box-only label(s) in a segmentation dataset. Ultralytics drops ALL masks when box/segment counts differ; converted to rectangle polygons | ✓ |
| warning | out_of_bounds | subj01_frame0000.jpg | obj#1: polygon leaves image; clipped | ✓ |
| warning | tiny_box | subj03_frame0000.jpg | obj#1: box 1.2x40.0px smaller than 2.0px; dropped | ✓ |
| info | bbox_polygon_mismatch | subj01_frame0000.jpg | obj#1: bbox differs from polygon extent; bbox recomputed | ✓ |
Sample images with labels
