[2007.09748] A Generic Visualization Approach for Convolutional Neural Networksopen searchopen navigation menucontact arXivsubscribe to arXiv mailings

Retrieval networks are essential for searching and indexing. Compared to classification networks, attention visualization for retrieval networks is hardly studied. We formulate attention visualization as a constrained optimization problem. We leverage the unit L2-Norm constraint as an attention filter (L2-CAF) to localize attention in both classification and retrieval networks. Unlike recent literature, our approach requires neither architectural changes nor fine-tuning. Thus, a pre-trained network's performance is never undermined L2-CAF is quantitatively evaluated using weakly supervised object localization. State-of-the-art results are achieved on classification networks. For retrieval networks, significant improvement margins are achieved over a Grad-CAM baseline. Qualitative evaluation demonstrates how the L2-CAF visualizes attention per frame for a recurrent retrieval network. Further ablation studies highlight the computational cost of our approach and compare L2-CAF with othe

2 mentions: @AkiraTOSEI@AkiraTOSEI
Date: 2020/09/14 11:22

Referring Tweets

@AkiraTOSEI t.co/csAemzB6aL Attentionベースの根拠可視化手法L2-CAFを提案。特徴量マップから検出のクラス以外のものを排除するフィルターを通すことで可視化する。Softmax出力から信号を得と単一モードの信号しか得られないが、複数モードを扱えるから有利とのこと。 t.co/ltSoFSt1BB
@AkiraTOSEI t.co/csAemzB6aL They proposed an Attention-based evidence visualization method, L2-CAF. The method is based on a filter that eliminates all features other than the target class from the feature map, and it has the advantage of being able to handle multiple modes. t.co/Y2E5grCfhu

Related Entries

Read more [1908.09231] Towards Unconstrained End-to-End Text Spottingcontact arXivarXiv Twitter
0 users, 2 mentions 2020/01/16 02:21
Read more [1911.09785] ReMixMatch: Semi-Supervised Learning with Distribution Alignment and Augmentation Ancho...
0 users, 2 mentions 2020/02/05 14:20
Read more [1912.04158] Learning a Neural 3D Texture Space from 2D Exemplarscontact arXivarXiv Twitter
0 users, 2 mentions 2020/03/26 11:21
Read more [1912.02424] Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training S...
0 users, 2 mentions 2020/07/30 12:53
Read more [1908.11512] Fast and Accurate Network Embeddings via Very Sparse Random Projectionopen searchopen n...
0 users, 2 mentions 2020/08/24 11:21