Table 2. An ablation test of Selfee (Self-supervised Features Extraction) training process on fly datasets.

Model	Pre-trained ResNet-50 with random projectors		Selfee		Selfee without CLD loss		Selfee without Turbo transformation
Evaluation	Mean F1 score	Mean AP	Mean F1 score	Mean AP	Mean F1 score	Mean AP	Mean F1 score	Mean AP
Replication 1	0.586	0.580	0.724	0.758	0.227	0.227	0.604	0.550
Replication 2	0.597	0.570	0.676	0.683	0.163	0.200	0.574	0.551
Replication 3	0.596	0.586	0.714	0.754	0.172	0.214	0.517	0.497
Best	0.597	0.586	0.724 *	0.758 *	0.227	0.227	0.604	0.551

^*Best results of different training setups under each evaluation metric are indicated in bold values.