ATOM: Accurate Tracking by Overlap Maximization

Abstract

While recent years have witnessed astonishing improvements in visual tracking robustness, the advancements in tracking accuracy have been limited. As the focus has been directed towards the development of powerful classifiers, the problem of accurate target state estimation has been largely overlooked. In fact, most trackers resort to a simple multi-scale search in order to estimate the target bounding box. We argue that this approach is fundamentally limited since target estimation is a complex task, requiring highlevel knowledge about the object.We address this problem by proposing a novel tracking architecture, consisting of dedicated target estimation and classification components. High level knowledge is incorporated into the target estimation through extensive offline learning. Our target estimation component is trained to predict the overlap between the target object and an estimated bounding box. By carefully integrating target-specific information, our approach achieves previously unseen bounding box accuracy. We further introduce a classification component that is trained online to guarantee high discriminative power in the presence of distractors. Our final tracking framework sets a new state-of-the-art on five challenging benchmarks. On the new large-scale Track-ingNet dataset, our tracker ATOM achieves a relative gain of 15% over the previous best approach, while running at over 30 FPS. Code and models are available at https: //github.com/visionml/pytracking. * Both authors contributed equally. ATOMDaSiamRPN UPDT

Learning Discriminative Model Prediction for Tracking

“…In this work, we introduce an alternative tracking architecture, trained in an end-to-end manner, that directly addresses all aforementioned limitations. In our design, we take inspiration from the discriminative learning procedures that have been successfully applied in recent trackers [26,7,4]. Our approach is based on a target model prediction network, which is derived from a discriminative learning loss by applying an iterative optimization procedure.…”

Learning the Model Update for Siamese Trackers

“…Finally the top-performance trackers are shown in Figure 6. We compare with trackers including DRT [37], DeepSTRCF [28], LSART [38], R MCPF [48], SRCT [25], CSRDCF [30], LADCF [44], MFT [22], UPDT [2] and ATOM [8] among others as in [22]. Among the top trackers, our approach achieves superior performance while maintaining a very high efficiency.…”

Visual Tracking by Means of Deep Reinforcement Learning and an Expert Demonstrator

“…A3CTD increases additionally the performance of A3CT, with an improvement of 1% in AO, 1.8 in SR 0.50 but with a loss of 0.7% in SR 0.75 . We perform worse than ATOM [6], however we remark that these results are obtained considering just 1782 of the 9335 sequences (19%) contained in the GOT-10k training set.…”