Active Learning for Human Pose Estimation

Abstract

Annotating human poses in realistic scenes is very time consuming, yet necessary for training human pose estimators. We propose to address this problem in an active learning framework, which alternates between requesting the most useful annotations among a large set of unlabelled images, and re-training the pose estimator. To this end, (1) we propose an uncertainty estimator specific for body joint predictions, which takes into account the spatial distribution of the responses of the current pose estimator on the unlabelled images; (2) we propose a dynamic combination of influence and uncertainty cues, where their weights vary during the active learning process according to the reliability of the current pose estimator; (3) we introduce a computer assisted annotation interface, which reduces the time necessary for a human annotator to click on a joint by discretizing the image into regions generated by the current pose estimator. Experiments using the MPII and LSP datasets with both simulated and real annotators show that (1) the proposed active selection scheme outperforms several baselines; (2) our computer-assisted interface can further reduce annotation effort; and (3) our technique can further improve the performance of a pose estimator even when starting from an already strong one.

Learning Loss for Active Learning

“…As the two approaches based on uncertainty and distribution are differently motivated, they are complementary to each other. Thus, a variety of hybrid strategies have been proposed [29,59,41,56] for their specific tasks.…”

VaB-AL: Incorporating Class Imbalance and Difficulty with Variational Bayes for Active Learning

“…Beyond them, hybrid approaches combine the best of both uncertainty-based and representation-based methods [34,50]. Some works focused on a specific task: for example person re-identification [29] and a human pose estimation [28].…”

Deep Active Learning For Human Pose Estimation Via Consistency Weighted Core-Set Approach

“…In addition to the random selection, we compare the performance of our method with some state-ofthe-art AL methods as well, which include uncertainty based LL [9] and representation based VAAL [3] and Core-set [1]. Moreover, the result of a hybrid AL method AALU [17] and a task-specific AL method for pose estimation AL4Pose [16] are also reported as competitive baselines. Implementation Detail.…”

“…Moreover, all the above methods [1][2][3][4][5][6][7][8][9][10][11][12] focus either on uncertainty or representativeness individually but lack the combination strategy. Although some hybrid approaches [13][14][15][16][17][18][19] have been proposed to address this limitation, it is still a core challenge to fuse the two complementary information properly and adaptively during data selection.…”