mAP@0.1:0.7: 47.3mAP@0.5: 37.3
Papers with Code · CC BY-SA 4.0
Zero-Shot Learning
Zero-shot learning (ZSL) is a model's ability to detect classes never seen during training. The condition is that the classes are not known during supervised learning.
Earlier work in zero-shot learning use attributes in a two-step approach to infer unknown classes. In the computer vision context, more recent advances learn mappings from image feature space to semantic space. Other approaches learn non-linear multimodal embeddings. In the modern NLP context, language models can be evaluated on downstream tasks without fine tuning.
Benchmark datasets for zero-shot learning include aPY, AwA, and CUB, among others.
( Image credit: Prototypical Networks for Few shot Learning in PyTorch )
Further readings:
논문
mAP@0.1:0.7: 43.5mAP@0.5: 33.9
mAP@0.1:0.7: 33.7mAP@0.5: 21.9
Harmonic Mean (10 unseen classes): 60.42Harmonic Mean (24 unseen classes): 44.50Random Split Harmonic Mean: 47.54
mAP@0.1:0.7: 49.2
mAP@0.1:0.7: 31.0mAP@0.5: 19.3
mAP@0.1:0.7: 36.4mAP@0.5: 28.8
mAP@0.5: 34.6mAP@0.1:0.7: 42.6mAP@0.1:0.5: 53.2
mAP@0.5: 36.9mAP@0.1:0.7: 46.4mAP@0.1:0.5: 56.9
mAP@0.5: 38.3mAP@0.1:0.7: 44.6mAP@0.1:0.5: 54.4
mAP@0.1:0.7: 76.9mAP@0.5: 66.3
Harmonic Mean (10 unseen classes): 57.40Harmonic Mean (24 unseen classes): 52.40
mAP@0.5: 41.8mAP@0.1:0.7: 50.6mAP@0.1:0.5: 61.5
mAP@0.5: 42.8mAP@0.1:0.7: 49.6mAP@0.1:0.5: 60.1
mAP@0.5: 40.0mAP@0.1:0.7: 47.0mAP@0.1:0.5: 57.4
Harmonic Mean (10 unseen classes): 54.94Harmonic Mean (24 unseen classes): 41.04
mAP@0.5: 37.1mAP@0.1:0.7: 46.2mAP@0.1:0.5: 57.2
mAP@0.5: 41.0mAP@0.1:0.7: 48.5mAP@0.1:0.5: 59.1
mAP@0.5: 33.7mAP@0.1:0.7: 41.9mAP@0.1:0.5: 51.6