1: \begin{abstract}

2: The training scheme of deep face recognition has greatly evolved in the past years, yet it encounters new challenges in the large-scale data situation where massive and diverse hard cases occur.

3: % problem

4: Especially in the range of low false accept rate (FAR), there are various hard cases in both positives (intra-class) and negatives (inter-class). 

5: % our objective

6: In this paper, we study how to make better use of these hard samples for improving the training.

7: % previous solution

8: The literature approaches this by margin-based formulation in either positive logit %(such as SphereFace, CosFace, ArcFace) 

9: or negative logits. %(such as MV-softmax, ArcNegFace, CurricularFace).

10: % their flaws

11: However, the correlation between hard positive and hard negative is overlooked, and so is the relation between the margins in positive and negative logits.

12: We find such correlation is significant, especially in the large-scale dataset, and one can take advantage from it to boost the training via relating the positive and negative margins for each training sample.

13: % our goal

14: To this end, we propose an explicit collaboration between positive and negative margins sample-wisely.

15: % our method

16: Given a batch of hard samples, a novel Negative-Positive Collaboration loss, named NPCFace, is formulated, which emphasizes the training on both negative and positive hard cases via the collaborative-margin mechanism in the softmax logits, and also brings better interpretation of negative-positive hardness correlation.

17: Besides, the emphasis is implemented with an improved formulation to achieve stable convergence and flexible parameter setting. 

18: % problem to cope with

19: % conclusion

20: We validate the effectiveness of our approach on various benchmarks of large-scale face recognition, and obtain advantageous results especially in the low FAR range.

21: \end{abstract}

22: