허재필 교수 연구실, ICCV 2021 국제학술대회 논문 2편 게재 승인 (노해찬 박사과정, 김태호 석박통합과정, 주원영 석사과정)
- 인공지능학과
- 조회수2266
- 2021-08-30
비주얼컴퓨팅연구실(지도교수: 허재필)의 논문 2편이 컴퓨터 비전 및 인공지능 분야의 Top-tier 학술대회인 IEEE/CVF International Conference on Computer Vision (ICCV) 2021 에 게재 승인되었습니다
논문 #1: “Product Quantizer Aware Inverted Index for Scalable Nearest Neighbor Search” (인공지능학과 박사과정 노해찬 학우, 인공지능학과 석박통합과정 김태호 학우)
논문 #2: “Collaborative Learning with Disentangled Features for Zero-shot Domain Adaptation” (인공지능학과 석사과정 주원영 학우)
“Product Quantizer Aware Inverted Index for Scalable Nearest Neighbor Search” 에서는 대용량 데이터 베이스에 대한 최근접 이웃 근사(Nearest Neighbor Approximation) 기술에 사용되는 역색인 (Inverted Indexing) 구조의 새로운 학습 방법을 제시하였습니다. 기존의 기술들은 탐색 속도의 복잡도를 줄이기 위한 역색인 구조와 속도 및 메모리 사용량을 줄이기 위한 손실 압축 기법을 동시에 사용하지만 각각의 기법은 독립적으로 학습되었습니다. 본 연구에서는 이 두 가지 기법을 공동 최적화 (Joint Optimization)를 통해 압축 기법의 왜곡 (Distortion) 을 줄이는 학습 방법을 제안하여 대용량 데이터 베이스에 대한 최근접 이웃 근사 기술 분야에서 가장 높은 성능을 달성하였습니다.
“Collaborative Learning with Disentangled Features for Zero-shot Domain Adaptation” 연구에서는 전이학습의 한 분야인 Zero-shot Domain Adaptation (ZSDA) 을 위한 새로운 프레임워크를 제시하였습니다. ZSDA는 타겟 도메인의 특정 클래스에 대한 데이터가 존재하지 않을 때, 다른 클래스들의 도메인 변화 (Domain Shift) 를 포착하여 도메인 적응을 시도하는 기술입니다. 제안하는 모델에서는 이미지에서 도메인 특징점과 의미론적 (Semantic) 특징점을 추출한 뒤, 두 특징점간의 협력적 학습과정 (Collaborative Learning) 을 통해 클래스를 예측하도록 설계하였습니다. 제안된 모델은 현재 ZSDA 분야에서 가장 높은 성능을 달성하였으며, 추후 Zero-shot Learning 및 도메인 적응 연구에 큰 도움이 될 것입니다.
[논문 #1 정보]
Product Quantizer Aware Inverted Index for Scalable Nearest Neighbor Search
Haechan Noh, Taeho Kim, and Jae-Pil Heo
IEEE/CVF International Conference on Computer Vision (ICCV), 2021
Abstract:
The inverted index is one of the most commonly used structures for non-exhaustive nearest neighbor search on large-scale datasets. It allows a significant factor of acceleration by a reduced number of distance computations with only a small fraction of the database. In particular, the inverted index enables the product quantization (PQ) to learn their codewords in the residual vector space. The quantization error of the PQ can be substantially improved in such combination since the residual vector space is much more quantization-friendly thanks to their compact distribution compared to the original data. In this paper, we first raise an unremarked but crucial question; why the inverted index and the product quantizer are optimized separately even though they are closely related? For instance, changes on the inverted index distort the whole residual vector space. To address the raised question, we suggest a joint optimization of the coarse and fine quantizers by substituting the original objective of the coarse quantizer to end-to-end quantization distortion. Moreover, our method is generic and applicable to different combinations of coarse and fine quantizers such as inverted multi-index and optimized PQ.
[논문 #2 정보]
Collaborative Learning with Disentangled Features for Zero-shot Domain Adaptation
Won Young Jhoo, and Jae-Pil Heo
IEEE/CVF International Conference on Computer Vision (ICCV), 2021
Abstract:
Typical domain adaptation techniques aim to transfer the knowledge learned from a label-rich source domain to a label-scarce target domain in the same label space. However, it is often hard to get even the unlabeled target domain data of a task of interest. In such a case, we can capture the domain shift between the source domain and target domain from an unseen task and transfer it to the task of interest, which is known as zero-shot domain adaptation (ZSDA). Most of existing state-of-the-art methods for ZSDA attempted to generate target domain data. However, training such generative models causes significant computational overhead and is hardly optimized. In this paper, we propose a novel ZSDA method that learns a task-agnostic domain shift by collaborative training of domain-invariant semantic features and task-invariant domain features via adversarial learning. Meanwhile, the spatial attention map is learned from disentangled feature representations to selectively emphasize the domain-specific salient parts of the domain-invariant features. Experimental results show that our ZSDA method achieves state-of-the-art performance on several benchmarks.