Scene Graph Prediction with Limited Labels

Visual knowledge bases such as Visual Genome power numerous applications in computer vision, including visual question answering and captioning, but suffer from sparse, incomplete relationships. All scene graph models to date are limited to training on a small set of visual relationships that have thousands of training labels each... Hiring human annotators is expensive, and using textual knowledge base completion methods are incompatible with visual data. In this paper, we introduce a semi-supervised method that assigns probabilistic relationship labels to a large number of unlabeled images using few labeled examples. We analyze visual relationships to suggest two types of image-agnostic features that are used to generate noisy heuristics, whose outputs are aggregated using a factor graph-based generative model. With as few as 10 labeled examples per relationship, the generative model creates enough training data to train any existing state-of-the-art scene graph model. We demonstrate that our method outperforms all baseline approaches on scene graph prediction by 5.16 recall@100 for PREDCLS. In our limited label setting, we define a complexity metric for relationships that serves as an indicator (R^2 = 0.778) for conditions under which our method succeeds over transfer learning, the de-facto approach for training with limited labels. read more

PDF Abstract ICCV 2019 PDF ICCV 2019 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Predicate Classification ImageCLEF-DA LimLabel (Categ. + Spat.) R@100 47.53 # 1
R@20 45.49 # 1
R@50 47.04 # 1
Scene Graph Detection ImageCLEF-DA LimLabel (Categ. + Spat.) R@100 19.28 # 1
R@50 18.69 # 1
Scene Graph Classification ImageCLEF-DA LimLabel (Categ. + Spat.) R@100 21.44 # 1
R@20 20.91 # 1
R@50 21.34 # 1
Scene Graph Detection Visual Genome LimLabel (Categ. + Spat.) R@100 8.64 # 1
R@20 4.04 # 1
R@50 6.75 # 1
Scene Graph Classification Visual Genome LimLabel (Categ. + Spat.) R@100 14.16 # 1
R@20 12.69 # 2
R@50 13.91 # 1
Predicate Classification Visual Genome LimLabel (Categ. + Spat.) R@100 28.53 # 1
R@20 24.72 # 1
R@50 27.76 # 1
Scene Graph Detection VRD LimLabel (Categ. + Spat.) R@20 17.67 # 1

Methods


No methods listed for this paper. Add relevant methods here