Computation time is an important performance characteristic of computer vision algorithms. The paper shows how existing (slow) binary decision algorithms can be approximated by a (fast) trained WaldBoost classifier. WaldBoost learning minimises the decision time of the classifier while guaranteeing predefined precision. We show that the WaldBoost algorithm together with bootstrapping is able to efficiently handle an effectively unlimited number of training examples provided by the implementation of the approximated algorithm. Two interest point detectors, the Hessian-Laplace and the Kadir-Brady saliency detectors, are emulated to demonstrate the approach. Experiments show that while the repeatability and matching scores are similar for the original and emulated algorithms, a 9-fold speed-up for the Hessian-Laplace detector and a 142-fold speed-up for the Kadir-Brady detector is achieved. For the Hessian-Laplace detector, the achieved speed is similar to SURF, a popular and very fast handcrafted modification of Hessian-Laplace; the WaldBoost emulator approximates the output of the Hessian-Laplace detector more precisely.

Learning Fast Emulators of Binary Decision Processes

Confluence is a novel non-Intersection over Union (IoU) alternative to Non-Maxima Suppression (NMS) in bounding box post-processing in object detection. It overcomes the inherent limitations of IoU-based NMS variants to provide a more stable, consistent predictor of bounding box clustering by using a normalized Manhattan Distance inspired proximity metric to represent bounding box clustering. Unlike Greedy and Soft NMS, it does not rely solely on classification confidence scores to select optimal bounding boxes, instead selecting the box which is closest to every other box within a given cluster and removing highly confluent neighboring boxes. Confluence is experimentally validated on the MS COCO and CrowdHuman benchmarks, improving Average Precision by up to 2.3-3.8% and Average Recall by up to 5.3-7.2% when compared against de-facto standard and state of the art NMS variants. Quantitative results are supported by extensive qualitative analysis and threshold sensitivity analysis experiments support the conclusion that Confluence is more robust than NMS variants. Confluence represents a paradigm shift in bounding box processing, with potential to replace IoU in bounding box regression processes.

Confluence: A Robust Non-IoU Alternative to Non-Maxima Suppression in Object Detection.

Detection of objects in images using statistical classifiers is a well studied and documented technique. Different applications of such detectors often require selection of the image position with the highest response of the detector—they perform non-maxima suppression. This article introduces the concept of early non-maxima sup- pression, which aims to reduce necessary computations by making the non-maxima suppression decision early based on incomplete information provided by a partially evalu- ated classifier. We show that the error of one such specu- lative decision with respect to a decision made based on response of the complete classifier can be estimated by collecting statistics on unlabeled data. The article then considers a sequential strategy of multiple early non- maxima suppression tests which follows the structure of soft-cascade detectors commonly used for object detection. We also show that an optimal (fastest for requested error rate) suppression strategy can be created by a novel variant of Wald's sequential probability ratio test (SPRT) which we call the conditioned SPRT (CSPRT). Experimental results show that the early non-maxima suppression sig- nificantly reduces amount of computation in the case of object localization while the error rates are limited to low predefined values. The proposed approach notably outper- forms the state-of-the-art detectors based on WaldBoost. The potential applications of the early non-maxima suppression approach are not limited to object localization and could be applied wherever the goal is to find the strongest response of a classifier among a set of classified samples.

Pavel Zemc ˇ ik

Papers

Speeding up pattern localization and other tasks