A Deep Convolutional Neural Network for Location Recognition and Geometry based Information

Bidoia, Francesco; Sabatelli, Matthia; Shantia, Amirhossein; Wiering, Marco A.; Schomaker, Lambert

doi:10.5220/0006542200270036

Cited by 5 publications

(6 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition to that, in order to establish the potential benefits that TL from ImageNet has over training a DCNN from scratch, we also report the results that have been obtained when training one DCNN with weights that have been initially sampled from a "He-Uniform" distribution [12]. Since we take advantage of work [4] we use the Inception-V3 architecture. We refer to it in all figures as Scratch-V3 and visualize it with a solid orange line.…”

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Deep Transfer Learning for Art Classification Problems

Sabatelli

Kestemont

Daelemans

et al. 2019

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

In this paper we investigate whether Deep Convolutional Neural Networks (DCNNs), which have obtained state of the art results on the ImageNet challenge, are able to perform equally well on three different art classification problems. In particular, we assess whether it is beneficial to fine tune the networks instead of just using them as off the shelf feature extractors for a separately trained softmax classifier. Our experiments show how the first approach yields significantly better results and allows the DCNNs to develop new selective attention mechanisms over the images, which provide powerful insights about which pixel regions allow the networks successfully tackle the proposed classification challenges. Furthermore, we also show how DCNNs, which have been fine tuned on a large artistic collection, outperform the same architectures which are pre-trained on the ImageNet dataset only, when it comes to the classification of heritage objects from a different dataset.

show abstract

Section: Resultsmentioning

confidence: 99%

“…For our experiments we use two datasets which come from two different heritage collections. The first one contains the largest number of samples and comes from the Rijksmuseum in Amsterdam 4 . On the other hand, our second 'Antwerp' dataset is much smaller.…”

Section: Datasets and Classification Challengesmentioning

confidence: 99%

Deep Transfer Learning for Art Classification Problems

Sabatelli

Kestemont

Daelemans

et al. 2019

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

show abstract

“…Previously, we transferred the knowledge from a traditional map to a stacked denoising autoencoder (SDA) in which the robot used grid mapping for training data, and could localize its position using a camera after training in a small environment [21]. Bidoia et al [40], used a semi-supervised approach to create a graph map using deep CNNs. QR codes were scattered in the environment while the robot randomly moved throughout the environment.…”

Section: Previous Workmentioning

confidence: 99%

Two-stage visual navigation by deep neural networks and multi-goal reinforcement learning

Shantia

Timmers

Chong

et al. 2021

Robotics and Autonomous Systems

Self Cite

View full text Add to dashboard Cite

show abstract

“…Previously, we transferred the knowledge from a traditional map to a stacked denoising autoencoder (SDA) in which the robot used grid mapping for training data, and could localize its position using a camera after training in a small environment (Shantia, Timmers, Schomaker and Wiering 2015). Bidoia et al (Bidoia et al 2018), used a semi-supervised approach to create a graph map using deep CNNs. QR codes were scattered in the environment while the robot randomly moved throughout the environment.…”

Section: Previous Workmentioning

confidence: 99%

“…When applied in multiple layers, it works as a scale, and shift invariant feature for the network. This property, although useful for object recognition, is detrimental for precise location estimation (Bidoia et al 2018). We would like the network to observe exactly where an edge was in the image and with what size, since this is important for localizing the agent.…”

Section: Convolutional Neural Networkmentioning

confidence: 99%

Machine Learning in Robotic Navigation: Deep Visual Localization and Adaptive Control

Shantia¹

View full text Add to dashboard Cite

Samenvatting 114Where is the robot? Humans have an excellent notion of their current surroundings. Even if somebody kidnaps us and puts us in an unknown location, we can describe the characteristics of this new place without much effort. Our knowledge and experience of our world allow us to identify shapes, geometry, and objects that are around us. Besides, we can correctly connect a series of observations from our senses to build a spatial map. We are so good at it that even without vision, and with some training, we can move around a block and find our starting position. We do get lost, but usually, with a hint of a landmark, a passerby, or our phone's GPS, we adjust our estimated location. We can program robots to partially perform the localization and mapping tasks by using a variety of sensors such as cameras and range Chapter

show abstract

A Deep Convolutional Neural Network for Location Recognition and Geometry based Information

Cited by 5 publications

References 11 publications

Deep Transfer Learning for Art Classification Problems

Deep Transfer Learning for Art Classification Problems

Two-stage visual navigation by deep neural networks and multi-goal reinforcement learning

Machine Learning in Robotic Navigation: Deep Visual Localization and Adaptive Control

Contact Info

Product

Resources

About