Salient Object Subitizing

By Jianming Zhang, Shugao Ma, Mehrnoosh Sameki, Stan Sclaroff, Margrit Betke, Zhe L. Lin, Xiaohui Shen, Brian L. Price, Radomír Mech. 2015

119 citations recorded by the source corpus.

We study the problem of salient object subitizing, i.e. predicting the existence and the number of salient objects in an image using holistic cues. This task is inspired by the ability of people to quickly and accurately identify the number of items within the subitizing range (1–4). To this end, we present a salient object subitizing image dataset of about 14 K everyday images which are annotated using an online crowdsourcing marketplace. We show that using an end-to-end trained convolutional neural network (CNN) model, we achieve prediction accuracy comparable to human performance in identifying images with zero or one salient object. For images with multiple salient objects, our model also provides significantly better than chance performance without requiring any localization process. Moreover, we propose a method to improve the training of the CNN subitizing model by leveraging synthetic images. In experiments, we demonstrate the accuracy and generalizability of our CNN subitizing model and its applications in salient object detection and image retrieval.

Subfields: Cognitive Neuroscience, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Computer Vision and Pattern Recognition

Citations in the corpus

Learning to decompose for object detection and instance segmentation (2015, background): “luating a further classiﬁer over possible bounding boxes (chosen based on the activation map). Other approaches try to directly count objects without explicitly addressing recognition or localization Zhang et al. (2015). Toward general computational savings, Ren et al. (2015), showed deep CNN featur”
Attend, Infer, Repeat: Fast Scene Understanding with Generative Models (2016, background): “By its nature AIR is also related to the following problems: counting [14, 27], pondering [2], and gradient estimation through renderers [15].”
Unconstrained Salient Object Detection via Proposal Subset Optimization (2016, methodology): “In [53], a salient object subitizing model is proposed to suppress the detections on background images that contain no salient object.”
Salient Object Detection: A Benchmark (2015, methodology): “[117] introduced a fast method for a more challenging task of counting (subitizing) salient objects in a scene.”
Salient Objects in Clutter: Bringing Salient Object Detection to the Foreground (2018, background): “, salient object subitizing [45], instance level salient object detection [29], weakly supervised based salient object detection [27], etc.”
Interpretable Visual Question Answering by Reasoning on Dependency Trees (2018, background): “[32] studied the problem of salient object subitizing,”
Multimodal Contrastive Training for Visual Representation Learning (2021, unknown): “Similarly, we also evaluate the Stock pretrained model on multiple downstream tasks including image tagging and cross-modal retrieval on Stock’s test set to demonstrate effectiveness of our method in a large-scale dataset setting.”
Salient Object Detection in the Deep Learning Era: An In-Depth Survey (2019, background): “[67] .”
MoVie: Revisiting Modulated Convolutions for Visual Counting and Beyond (2021, background): “It is specialized because the number of objects is usually limited to be small ( e.g . up to 4 (Zhang et al., 2015)).”
Object Counting and Instance Segmentation With Image-Level Supervision (2019, background): “Reducing object count supervision for salient object subitizing was investigated in [31].”