Introduces BIP framework and GapGen generator to allocate and synthesize millions of non-colliding virtual face identities within gaps of the real face manifold.
hub
Learning Face Representation from Scratch
17 Pith papers cite this work. Polarity classification is still indexing.
17
Pith papers citing it
abstract
Pushing by big data and deep convolutional neural network (CNN), the performance of face recognition is becoming comparable to human. Using private large scale training datasets, several groups achieve very high performance on LFW, i.e., 97% to 99%. While there are many open source implementations of CNN, none of large scale face dataset is publicly available. The current situation in the field of face recognition is that data is more important than algorithm. To solve this problem, this paper proposes a semi-automatical way to collect face images from Internet and builds a large scale dataset containing about 10,000 subjects and 500,000 images, called CASIAWebFace. Based on the database, we use a 11-layer CNN to learn discriminative representation and obtain state-of-theart accuracy on LFW and YTF. The publication of CASIAWebFace will attract more research groups entering this field and accelerate the development of face recognition in the wild.
hub tools
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
Face image quality is quantified as the Euclidean distance between embeddings from a pre-trained face recognition model and its pruned version, achieving competitive or superior results without training or supervision.
StyleID supplies human-perception-aligned benchmarks and fine-tuned encoders that improve facial identity recognition robustness across stylization types and strengths.
A coupled RGB diffusion model conditioned on MLLM hidden-state embeddings performs reference-free single-image facial demorphing and outperforms latent-space and ViT baselines in ablations.
The paper shows that multiple-identity image attacks succeed due to modest angular separation between matching (~90°) and non-matching (40-60°) face representations, with image morphing and representation inversion realizing effective attacks that transfer across comparators.
BID-LoRA uses bi-directional low-rank adapters with retain/new/unlearn pathways and escape unlearning to enable continual learning and unlearning while minimizing knowledge leakage and parameter updates.
DSCL disentangles gaze features via Jacobian regularization into subspaces and applies ordinal contrastive learning to reach competitive in-domain and cross-domain performance with 5-20% labeled data.
Common feature learning transforms features from heterogeneous teacher networks into a shared space so a student model can imitate them all and outperform individual teachers without annotations.
Face segmentation for background removal systematically impacts both face recognition performance and morphing attack detection in unconstrained scenarios.
A reinforcement learning approach adapts general generative models to produce synthetic data that boosts identity recognition accuracy and generalization under privacy constraints.
Simple affine transformations align face embeddings across different DNN models, substantially improving cross-model identification and verification performance.
Deep face models trained on MS-Celeb-1M and fine-tuned on VGGFace2 achieve state-of-the-art accuracies on SCFace and ICB-RW low-resolution benchmarks without using any of their training data by leveraging appearance variety, resolution distribution, resolution matching, and probe information content
A pairwise-augmented loss on CNNs is reported to deliver state-of-the-art accuracy on primate face classification, verification, closed-set and open-set identification for two species.
A lightweight hybrid CNN-Transformer framework for heterogeneous face recognition achieves competitive performance on cross-spectral benchmarks and standard RGB tasks using contrastive alignment and distillation.
NTGA is the first clean-label generalization attack under black-box settings but is vulnerable to adversarial training and image transformations, with newer attacks outperforming it.
Slim-Net uses stacked Slim Modules of depthwise separable convolutions to predict face attributes on CelebA at 91.24% accuracy with at least 25 times fewer parameters than comparable models.
citing papers explorer
No citing papers match the current filters.