Bloom-Counter
An open-source flower-counting tool based on color segmentation. It detects flowers using HSV threshold segmentation and connected-component analysis, and includes a simulator that generates test images as well as scripts for performance evaluation.
Timeline (last 90 days)
In the Bloom-Counter article, the reviewer pointed out that the authors summarized the overexposure condition as essentially unaffected, when in fact the 15 false positives (precision 0.933) were bleached leaf fragments disguised as buds, and that brightening to create the target and darkening to lose the target are two different failure modes.
Bloom-Counter's README claims a 99.7% detection recall on 6 simulated photos and 357 flowers.
The flowers generated by the simulator are tuned to fall within the detector's HSV petal range, so the 99.7% recall does not prove the detector's effectiveness in real scenarios.
The simulator's _place function uses rejection sampling to ensure that no two flowers ever overlap.
The scoring formula in demo.py is (detected buds + detected budlets) / (ground-truth buds + budlets), counting only totals, with no per-target matching and no breakdown by category.
The acceptance metrics of the counting pipeline must be per-category, per-target matched precision/recall, plus an explicit error for the bud-to-flower ratio; looking only at totals masks structural errors.
All Facts (6)
In the Bloom-Counter article, the reviewer pointed out that the authors summarized the overexposure condition as essentially unaffected, when in fact the 15 false positives (precision 0.933) were bleached leaf fragments disguised as buds, and that brightening to create the target and darkening to lose the target are two different failure modes.
50%UnverifiedBloom-Counter's README claims a 99.7% detection recall on 6 simulated photos and 357 flowers.
50%UnverifiedThe flowers generated by the simulator are tuned to fall within the detector's HSV petal range, so the 99.7% recall does not prove the detector's effectiveness in real scenarios.
50%UnverifiedThe simulator's _place function uses rejection sampling to ensure that no two flowers ever overlap.
50%UnverifiedThe scoring formula in demo.py is (detected buds + detected budlets) / (ground-truth buds + budlets), counting only totals, with no per-target matching and no breakdown by category.
50%UnverifiedThe acceptance metrics of the counting pipeline must be per-category, per-target matched precision/recall, plus an explicit error for the bud-to-flower ratio; looking only at totals masks structural errors.
50%