Precise Detection in Densely Packed Scenes

Goldman, Eran; Herzig, Roei; Eisenschtat, Aviv; Goldberger, Jacob; Hassner, Tal

doi:10.1109/cvpr.2019.00537

Cited by 182 publications

(146 citation statements)

References 46 publications

Supporting

Mentioning

145

Contrasting

Unclassified

Order By: Relevance

“…Object counting methods aim to predict the total number of objects in different categories existing in images, such as pedestrian counting [20,26,46,50], vehicle counting [12,49], goods counting [11,21] and general object counting [2,3,18]. In [2], the regression-based common object counting with image-level and instance-level supervision is investigated.…”

Section: Object Counting Algorithmsmentioning

confidence: 99%

“…The image-level counting strategy directly estimate the global count of objects without providing their location information [20,26,46,50]. The instance-level counting strategy predict an accurate number of objects with their location information (e.g., center point or bounding box) [3,11,12,21,43,49]. The object is represented by proposals with global appearance information in anchor based methods, it is not flexible to design different kinds of anchors because of large scale variation in drone based scenes.…”

Section: Object Counting Algorithmsmentioning

confidence: 99%

See 1 more Smart Citation

Guided Attention Network for Object Detection and Counting on Drones

Cai

Zhang

et al. 2020

Proceedings of the 28th ACM International Conference on Multimedia

View full text Add to dashboard Cite

Object detection and counting are related but challenging problems, especially for drone based scenes with small objects and cluttered background. In this paper, we propose a new Guided Attention network (GAnet) to deal with both object detection and counting tasks based on the feature pyramid. Different from the previous methods relying on unsupervised attention modules, we fuse different scales of feature maps by using the proposed weakly-supervised Background Attention (BA) between the background and objects for more semantic feature representation. Then, the Foreground Attention (FA) module is developed to consider both global and local appearance of the object to facilitate accurate localization. Moreover, the new data argumentation strategy is designed to train a robust model in the drone based scenes with various illumination conditions. Extensive experiments on three challenging benchmarks (i.e., UAVDT, CARPK and PUCPR+) show the state-of-the-art detection and counting performance of the proposed method compared with existing methods. Code can be found at https://isrc.iscas.ac.cn/gitlab/research/ganet.

show abstract

Section: Object Counting Algorithmsmentioning

confidence: 99%

Section: Object Counting Algorithmsmentioning

confidence: 99%

Guided Attention Network for Object Detection and Counting on Drones

Cai

Zhang

et al. 2020

Proceedings of the 28th ACM International Conference on Multimedia

View full text Add to dashboard Cite

show abstract

“…For dense object detection, different approaches have been used to deal with the problem of overlapping bounding boxes. One approach, discussed in [6], is to add an extra layer to the network (a soft intersection-over-union layer) to provide information on the quality of the detection boxes.…”

Section: Related Workmentioning

confidence: 99%

Detection of Dense, Overlapping, Geometric Objects

Peskin¹,

Wilthan²,

Majurski³

2020

IJAIA

View full text Add to dashboard Cite

Using a unique data collection, we are able to study the detection of dense geometric objects in image data where object density, clarity, and size vary. The data is a large set of black and white images of scatterplots, taken from journals reporting thermophysical property data of metal systems, whose plot points are represented primarily by circles, triangles, and squares. We built a highly accurate single class U-Net convolutional neural network model to identify 97 % of image objects in a defined set of test images, locating the centers of the objects to within a few pixels of the correct locations. We found an optimal way in which to mark our training data masks to achieve this level of accuracy. The optimal markings for object classification, however, required more information in the masks to identify particular types of geometries. We show a range of different patterns used to mark the training data masks, and how they help or hurt our dual goals of location and classification. Altering the annotations in the segmentation masks can increase both the accuracy of object classification and localization on the plots, more than other factors such as adding loss terms to the network calculations. However, localization of the plot points and classification of the geometric objects require different optimal training data.

show abstract

“… Name # Classes # Instances # Images GTIN Product data Citation Holoselecta 109 10035 295 Yes Yes, e.g. nutrients/price/brand [2] Grozi-3.2K 3235 3235 3235 + 680 No No [3] Grozi120 120 120 720 + 4973 No No [4] SKU110K 110,712 ∼1.74 * 10^6 11,762 No No [5] …”

Section: Data Descriptionmentioning

confidence: 99%

HoloSelecta dataset: 10’035 GTIN-labelled product instances in vending machines for object detection of packaged products in retail environments

Fuchs

Grundmann

Haldimann³

et al. 2020

Data in Brief

View full text Add to dashboard Cite

To assess the potential of current neural network architectures to reliably identify packaged products within a retail environment, we created an open-source dataset of 295 shelf images of vending machines with 10’035 labelled instances of 109 products. The dataset contains photos of vending machines by the provider Selecta, the largest European operator of vending machines. The vending machines are a mix of machines in public and private office spaces. The vending machines contain food as well as beverage products. The product instances in the vending machine images are labelled with bounding boxes, where a bounding box encapsulates the entire product with as little overlap as possible. The labels corresponding to the bounding box consist of a structured, human-readable labels including brand, product name and size as well as the GTIN of the product. The GTIN is the global standard to identify products in the retail environment and therefore increases the value as a dataset for the retail industry. Contrary to typical object detection datasets that choose labels at a higher level such as a can or bottle for a much wider variety of objects, this dataset chooses a far more detailed label that depends less on the shape but rather on the exact design of the product. The dataset falls into the category of object detection datasets with a large number of objects, which next to the GTIN label, represents a main differentiator of the dataset to other object detection datasets.

show abstract

Precise Detection in Densely Packed Scenes

Cited by 182 publications

References 46 publications

Guided Attention Network for Object Detection and Counting on Drones

Guided Attention Network for Object Detection and Counting on Drones

Detection of Dense, Overlapping, Geometric Objects

HoloSelecta dataset: 10’035 GTIN-labelled product instances in vending machines for object detection of packaged products in retail environments

Contact Info

Product

Resources

About