Dataset quality report

E:\Sythetic_Data\tool-detector\synthetic_raw

✓Passed · 0 errors · 0 warnings
1500
Images
3177
Boxes
2.12
Boxes per image
0
Empty images
0
Errors
0
Warnings

Checks

ResultCheckWhat it meansFound
✓ OKmissing_labelImage has no label file0
✓ OKorphan_labelLabel file has no image0
✓ OKunreadable_imageImage file can't be opened0
✓ OKmalformed_lineLabel line is not 'class x y w h' with numbers0
✓ OKbad_class_idClass id is not in the class list0
✓ OKout_of_boundsBox goes outside the image (coordinates outside 0-1)0
✓ OKzero_size_boxBox has zero or negative width or height0
✓ OKmissing_classA class has no boxes at all0
✓ OKempty_labelImage has no boxes (fine if it's a deliberate background image)0
✓ OKtiny_boxBox is very small and hard to learn from0
✓ OKduplicate_boxTwo boxes of the same class almost completely overlap0
✓ OKclass_imbalanceLargest class has many more boxes than the smallest0
✓ OKno_class_namesNo classes.txt found, class names are guessed0

Boxes per class

Balanced classes train more evenly. Limit before a warning: 3× between largest and smallest.

wrench740screwdriver999hammer728pliers710
Table view
ClassBoxesShare
wrench74023.3%
screwdriver99931.4%
hammer72822.9%
pliers71022.3%

Boxes per image

How many objects each image holds.

0250500750100001234

Box size

Box side length as a share of the image (√ of box area).

025050075010000%10%20%30%40%50%60%70%80%90%

Where boxes appear in the frame

Darker = more box centres. One hot spot in the middle means objects are always centred, which can hurt detection near the edges.

Sample images

Images with issues first (marked ⚠), then an even spread across the dataset. Resolutions: 1024x768 (902), 768x1024 (598).

All issues

No issues found.