Occluded Pedestrian Detection with Visible IoU and Box Sign Predictor

Abstract

Training a robust classifier and an accurate box regressor are difficult for occluded pedestrian detection. Traditionally adopted Intersection over Union (IoU) measurement does not consider the occluded region of the object and leads to improper training samples. To address such issue, a modification called visible IoU is proposed in this paper to explicitly incorporate the visible ratio in selecting samples. Then a newly designed box sign predictor is placed in parallel with box regressor to separately predict the moving direction of training samples. It leads to higher localization accuracy by introducing sign prediction loss during training and sign refining in testing. Following these novelties, we obtain state-of-the-art performance on CityPersons benchmark for occluded pedestrian detection.

PGNet: Pipeline Guidance for Human Key-Point Detection

“…Despite the detection network frameworks being different, the regression calculation logic for predicting the borders to locate the borders in the target object is the same. Shengkai et al [36] propose IoU-balanced loss functions that consist of IoUbalanced classification loss and IoU-balanced localization loss to solve poor localization accuracy, IoU is an important indicator for neural networks to measure between ground truths and predicted images. In object detection of a bounding box, the object being detected is the minimum value of the rectangular border through multiple iterations.…”

“…Despite the detection network frameworks being different, the regression calculation logic for predicting the borders to locate the borders in the target object is the same. Shengkai et al [36] propose IoU-balanced loss functions that consist of IoU-balanced classification loss and IoU-balanced localization loss to solve poor localization accuracy, and this is harmful for accurate localization. [37] proposed visible IoU to explicitly incorporate the visible ratio in selecting samples, which included a box regressor to separately predict the moving direction of training samples.…”

“…Both these losses can make key-point detectors more powerful for accurate localization. According to Equation 1, IoU loss can be defined such that Because IoU loss can only be effective when the bounding box is reattached during the training process, there are steps where gradient optimization cannot be performed without any coincidence [36]. To overcome the disadvantage that the border boxes must have coincidence to take advantage of IoU losses, GIoU was proposed.…”

Pose-Guided Part-Based Adaptive Pyramid Features for Occluded Person Reidentification

“…Each landmark contains its coordinates and confidence score. Inspired by [34], we employ a sigmoid form of decay function to better distinguish between occluded regions and nonoccluded regions. e decay function increases with the increase in the confidence of the landmark, and the sigmoid form is defined as follows:…”

Fast and accurate vision-based pedestrian detection

“…GDFL [141] proposes an attention mechanism to encode fine-grained pedestrian masks into feature maps in order to build more discriminative convolutional features for occluded pedestrians. IoU vis +Sign [158] proposes a visible IoU (Intersection over Union) to explicitly integrate visible ratio when selecting samples during the training phase, which improves the localization accuracy for occluded pedestrians. In TFAN+TDEM+PRM [159], a tube feature aggregation network is developed to exploit local temporal context of pedestrians from video to learn more robust pedestrian detector for severely occluded cases.…”