Machine Learning for Aerial Image Labeling

By Geoffrey E. Hinton, Volodymyr Mnih. 2013

911 citations recorded by the source corpus.

Subfields: 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 Computer Vision and Pattern Recognition

Citations in the corpus

Playing Atari with Deep Reinforcement Learning (2013, background): “ on the quality of the feature representation. Recent advances in deep learning have made it possible to extract high-level features from raw sensory data, leading to breakthroughs in computer vision [11, 22, 16] and speech recognition [6, 7]. These methods utilise a range of neural network architec”
Exploiting Cyclic Symmetry in Convolutional Neural Networks (2016, methodology): “onsists of 1500 1500 aerial images of the Boston area, with each image covering an area of 2.25 square kilometers. The dataset features 151 images, with pixelwise annotations of buildings. Following (Mnih, 2013), it was split into a training set of 137 images, a validation set of 4 images and a test”
DeepGlobe 2018: A Challenge to Parse the Earth through Satellite Images (2018, background): “Some of these can be enumerated as the TorontoCity[54] dataset, the ISPRS 2D semantic labeling dataset [3], the Mnih dataset [39], the SpaceNet dataset [2] and the ISPRS Benchmark for Multi-Platform Photogrammetry [4].”
RS-Mamba for Large Remote Sensing Image Dense Prediction (2024, methodology): “The Massachusetts [38] Road dataset incorporates 1171 aerial photographs from Massachusetts, with each image measuring 1500 × 1500 pixels and encompassing 2.25 km 2 .”
Learning Aerial Image Segmentation From Online Maps (2017, background): “scale OSM labels can be fine-tuned to achieve more accurate results for labeling remote sensing images [49], [50].”
SpaceNet: A Remote Sensing Dataset and Challenge Series (2018, background): “m aerial 4-channel imagery, and ˘700 km2 of coverage; building and roads are labeled at high ﬁdelity (among other items), but the data has yet to be publicly released. The Massachusetts Roads Dataset [9] contains 3-channel imagery at 1 meter resolution, and 2600 km2 of coverage; the imagery and labe”
Very Deep Convolutional Neural Networks for Complex Land Cover Mapping Using Multispectral Remote Sensing Imagery (2018, background): “In the patch-based CNN, a particular class label is assigned to the given patch when a small rectangle in the center of that patch completely covers a single object [42].”
Classification for High Resolution Remote Sensing Imagery Using a Fully Convolutional Network (2017, methodology): “The output number (channels) of the so in this paper, it is 12 for 12-class classiﬁcations).”
Convolutional Neural Networks for Large-Scale Remote-Sensing Image Classification (2017, positive): “For example, convolutional networks have fewer connections than multi-layer perceptrons but perform better in practice for visual tasks [1], and Mnih [29] showed that adding too many layers to a network resulted in poorer results.”
clDice - a Novel Topology-Preserving Loss Function for Tubular Structure Segmentation (2020, methodology): “In 2D, we evaluate on the DRIVE retina dataset [43], the Massachusetts Roads dataset [28] and the CREMI neuron dataset [12].”
Land-cover classification with high-resolution remote sensing images using transferable deep models (2018, background): “ed land-use classication approaches [44{46]. However, the geographic areas covered by most of existing landuse datasets [44,46,47] donot exceed 10km2 and somewhat similar in geographic distributions [48]. The lack of variations in geographic distributions of annotated HRRS images may cause overtting”
Topology-Preserving Deep Image Segmentation (2019, methodology): “3 Experiments We evaluate our method on six natural and biomedical datasets: CREMI6, ISBI12 [4], ISBI13 [3], CrackTree [41], Road [24] and DRIVE [34].”
Satlas: A Large-Scale, Multi-Task Dataset for Remote Sensing Image Understanding (2022, background): “The other three are the Massachusetts Buildings and Massachusetts Roads datasets [39], which involve semantic segmentation, and the Airbus Ships [2] dataset, which involves instance segmentation.”
Fully Convolutional Networks for Dense Semantic Labelling of High-Resolution Aerial Imagery (2016, background): “Learning would need to take into account noise in these less-accurate labels [19].”
Can semantic labeling methods generalize to any city? the inria aerial image labeling benchmark (2017, methodology): “With the goal of comparing classiﬁcation methods over large areas, Mnih [2] created building and road classiﬁcation datasets over Massachusetts, covering 340 km 2 and 2600 km 2 respectively.”