Training results: result_no_fp

[MateoLostanlen]

Evaluation results on test set

Metric	main	current	Δ
mAP@50	0.7556	0.6683	-0.0873
mAP@50-95	0.4802	0.4271	-0.0531
Precision	0.7886	0.7020	-0.0866
Recall	0.7361	0.6910	-0.0451

Sequential evaluation on test (optimized params: nb_frames=8, conf=0.4)

TP=113 FN=4 FP=12 TN=111

Metric	main	current	Δ
Recall (TP rate)	0.9597	0.9658	+0.0061
FPR	0.1409	0.0976	-0.0433
F1	0.9137	0.9339	+0.0202
Precision	0.8720	0.9040	+0.0320

Engine parameter optimization (top 10 on val)

nb_frames	conf	TP	FN	FP	TN	Recall	FPR	F1
8	0.4	90	13	11	113	0.8738	0.0887	0.8824
7	0.4	91	12	17	107	0.8835	0.1371	0.8626
8	0.35	93	10	20	104	0.9029	0.1613	0.8611
6	0.35	94	9	24	100	0.9126	0.1935	0.8507
7	0.35	94	9	24	100	0.9126	0.1935	0.8507
8	0.2	98	5	30	94	0.9515	0.2419	0.8485
5	0.35	95	8	26	98	0.9223	0.2097	0.8482
8	0.3	95	8	26	98	0.9223	0.2097	0.8482
6	0.4	91	12	21	103	0.8835	0.1694	0.8465
8	0.25	96	7	28	96	0.932	0.2258	0.8458

(vs last committed results on main)
Branch: result_no_fp | Test dataset: pyronear/pyro-dataset @ v2.1.0