Training results: result_ds_3.0.0_yolo26

[MateoLostanlen]

Evaluation results on test set

Metric	main	current	Δ
mAP@50	0.7556	0.7441	-0.0114
mAP@50-95	0.4802	0.4256	-0.0546
Precision	0.7886	0.7625	-0.0262
Recall	0.7361	0.6820	-0.0541

Sequential evaluation on test (optimized params: nb_frames=7, conf=0.3)

TP=114 FN=5 FP=4 TN=121

Metric	main	current	Δ
Recall (TP rate)	0.9597	0.9580	-0.0017
FPR	0.1409	0.0320	-0.1089
F1	0.9137	0.9620	+0.0483
Precision	0.8720	0.9661	+0.0941

Engine parameter optimization (top 10 on val)

nb_frames	conf	TP	FN	FP	TN	Recall	FPR	F1
7	0.3	103	8	6	125	0.9279	0.0458	0.9364
8	0.25	103	8	6	125	0.9279	0.0458	0.9364
6	0.25	104	7	8	123	0.9369	0.0611	0.9327
5	0.3	104	7	8	123	0.9369	0.0611	0.9327
7	0.25	104	7	8	123	0.9369	0.0611	0.9327
8	0.2	103	8	8	123	0.9279	0.0611	0.9279
6	0.3	102	9	7	124	0.9189	0.0534	0.9273
8	0.15	105	6	11	120	0.9459	0.084	0.9251
5	0.35	101	10	7	124	0.9099	0.0534	0.9224
7	0.35	100	11	6	125	0.9009	0.0458	0.9217

(vs last committed results on main)
Branch: result_ds_3.0.0_yolo26 | Test dataset: pyronear/pyro-dataset @ v3.0.0