Training results: result_2.3.0

[MateoLostanlen]

Evaluation results on test set

Metric	main	current	Δ
mAP@50	0.7556	0.7466	-0.0090
mAP@50-95	0.4802	0.4781	-0.0020
Precision	0.7886	0.7973	+0.0087
Recall	0.7361	0.7164	-0.0196

Sequential evaluation on test (optimized params: nb_frames=8, conf=0.25)

TP=114 FN=3 FP=8 TN=115

Metric	main	current	Δ
Recall (TP rate)	0.9597	0.9744	+0.0147
FPR	0.1409	0.0650	-0.0759
F1	0.9137	0.9540	+0.0403
Precision	0.8720	0.9344	+0.0624

Engine parameter optimization (top 10 on val)

nb_frames	conf	TP	FN	FP	TN	Recall	FPR	F1
8	0.25	97	6	8	116	0.9417	0.0645	0.9327
8	0.3	95	8	6	118	0.9223	0.0484	0.9314
8	0.4	91	12	2	122	0.8835	0.0161	0.9286
8	0.15	98	5	11	113	0.9515	0.0887	0.9245
7	0.4	91	12	3	121	0.8835	0.0242	0.9239
6	0.25	97	6	10	114	0.9417	0.0806	0.9238
5	0.3	97	6	10	114	0.9417	0.0806	0.9238
7	0.2	97	6	10	114	0.9417	0.0806	0.9238
8	0.2	97	6	10	114	0.9417	0.0806	0.9238
7	0.25	97	6	10	114	0.9417	0.0806	0.9238

(vs last committed results on main)
Branch: result_2.3.0 | Test dataset: pyronear/pyro-dataset @ v2.2.0