Human-centered deep compositional model for handling occlusions

Gregor Koporec (Avtor), Janez Perš (Avtor)

Povzetek

Despite their powerful discriminative abilities, Convolutional Neural Networks (CNNs) lack the properties of generative models. This leads to a decreased performance in environments where objects are poorly visible. Solving such a problem by adding more training samples can quickly lead to a combinatorial explosion, therefore the underlying architecture has to be changed instead. This work proposes a Human-Centered Deep Compositional model (HCDC) that combines low-level visual discrimination of a CNN and the high-level reasoning of a Hierarchical Compositional model (HCM). Defined as a transparent model, it can be optimized to real-world environments by adding compactly encoded domain knowledge from human studies and physical laws. The new FridgeNetv2 dataset and a mixture of publicly available datasets are used as a benchmark. The experimental results show the proposed model is explainable, has higher discriminative and generative power, and better handles the occlusion than the current state-of-the-art Mask-RCNN in instance segmentation tasks.

Ključne besede

računalniški vid;globoko učenje;konvolucijske nevronske mreže;hierarhični kompozicionalni model;zakrivanje;interpretabilnost;poznavanje področja;computer vision;deep learning;convolutional neural networks;hierarchical compositonal model;occlusion;discriminability;generalizability;interpretability;domain knowledge;

Podatki

Jezik:	Angleški jezik
Leto izida:	2023
Tipologija:	1.01 - Izvirni znanstveni članek
Organizacija:	UL FE - Fakulteta za elektrotehniko
UDK:	004
COBISS:	142438403
ISSN:	0031-3203
Št. ogledov:	18
Št. prenosov:	5
Ocena:	0 (0 glasov)
Metapodatki:

Ostali podatki

Sekundarni jezik:	Slovenski jezik
Sekundarne ključne besede:	računalniški vid;globoko učenje;konvolucijske nevronske mreže;hierarhični kompozicionalni model;zakrivanje;interpretabilnost;poznavanje področja;
Vrsta dela (COBISS):	Članek v reviji
Strani:	str. 1-14
Zvezek:	ǂVol. ǂ138, [article no.] ǂ109397
Čas izdaje:	2023
DOI:	10.1016/j.patcog.2023.109397
ID:	19861989