“Danbooru2020 Zero-Shot Anime Character Identification Dataset (ZACI-20)”, Kosuke Akimoto2021-02-06 (Danbooru AI; backlinks)⁠:

Danbooru2020 Zero-shot Anime Character Identification Dataset (ZACI-20): The goal of this dataset is creating human-level character identification models which do not require retraining on novel characters. The dataset is derived from Danbooru2020 dataset.

Features:

• Large-scale: 1.45M images of 39K characters (train dataset).

• Designed for zero-shot setting: characters in the test dataset do not appear in the train dataset, allowing us to test model performance on novel characters.

• Human-annotated test dataset:

  • Image pairs with erroneous face detection or duplicate images are manually removed.

  • We can compare model performance to human performance.

Benchmarks:

model name	FPR (%)	FNR (%)	EER (%)	note
Human	1.59	13.9	N/A	by kosuke1701
ResNet-152	2.40	13.9	8.89	w/ RandAugment, Contrastive loss. 0206_resnet152 by kosuke1701
SE-ResNet-152	2.43	13.9	8.15	w/ RandAug, Contrastive loss. 0206_seresnet152 by kosuke1701
ResNet-18	5.08	13.9	9.59	w/ RandAug, Contrastive loss. 0206_resnet18 by kosuke1701

Backlinks: