Sketch-a-Net: A Deep Neural Network that Beats Humans

By Qian Yu, Yongxin Yang, Feng Liu, Yi-Zhe Song, Tao A. Xiang, Timothy M. Hospedales. 2016

341 citations recorded by the source corpus.

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

ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness (2018, background): “, 2017; Brendel & Bethge, 2019), but they have much more difficulties recognising objects with missing texture information (Ballester & de Araújo, 2016; Yu et al., 2017).”
Deeper, Broader and Artier Domain Generalization (2017, neutral): “A more compelling motivation is domains where the total available images is fundamentally constrained, such as for particular styles of art [5, 29], and sketches [33, 6, 34, 26].”
PointCNN: Convolution On X-Transformed Points (2018, methodology): “Method TU-Berlin Quick Draw Sketch-a-Net [57] 77.95 - AlexNet [23] 68.60 - PointNet++ [35] 66 To verify this, we show T-SNE visualization of F o , F ∗ and F X of 15 randomly picked representative points from the ModelNet40 dataset in Figure 5, each with one color, and consistent in the sub-ﬁgures.”
More Photos are All You Need: Semi-Supervised Learning for Fine-Grained Sketch Based Image Retrieval (2021, methodology): “Handling Data-Scarcity for FG-SBIR: Earlier works have tried resolving the lack of instance-level photo-sketch paired data, by using edge-maps for training [30] or synthetic sketch stroke deformation [49, 48] for data augmentation.”
Semantic-Aware Knowledge Preservation for Zero-Shot Sketch-Based Image Retrieval (2019, background): “opment of deep learning, researchers have introduced deep neural networks into this field [22, 45, 21, 43, 37, 33, 30, 44, 42, 39].”
Sketch Less for More: On-the-Fly Fine-Grained Sketch-Based Image Retrieval (2020, neutral): “Secondly, although state-of-the-art vision systems are good at recognising badly drawn sketches [36, 50], users who perceive themselves as someone who “can’t sketch” worry about getting details wrong and receiving inaccurate results.”
Fine-Grained Image Retrieval: the Text/Sketch Input Dilemma (2017, methodology): “To obtain the multi-modal representations for this multi-view CCA framework, Sketch-A-Net [34] features (pool5 layer) are extracted for photo edges and sketches, respectively, while bag-of-words are applied to encode the text description.”
Partially Does It: Towards Scene-Level FG-SBIR with Partial Input (2022, methodology): “We compare against (i) existing state-of-the-art (SOTA) methods: Triplet-SN [79] employs Sketch-A-Net [80] backbone trained using triplet loss.”
Sketching without Worrying: Noise-Tolerant Sketch-Based Image Retrieval (2022, background): “That is, instead of costly operation like sketch deformation [59] or unfaithful approximation like edge/contour-map as soft ground-truths [10], users can effortlessly generate n most representative sub-sets to augment training for many downstream tasks.”
A Zero-Shot Framework for Sketch-based Image Retrieval (2018, methodology): “Yu et.al [24] were the ﬁrst to use Convolutional Neural Networks (CNN) for the sketch classiﬁcation task.”
StyleMeUp: Towards Style-Agnostic Sketch-Based Image Retrieval (2021, neutral): “81 % against Sketch-A-Net’s [61] 68 .”
Frankenstein: Learning Deep Face Representations Using Small Data (2016, neutral): “For example in the case of CNNs, training with data augmentation in the form of specifically designed deformation noise is important to obtain good recognition performance [64]; and in the case of de-noising auto encoders, training on images with noise, corruption and artefacts has Fig.”
SketchMate: Deep Hashing for Million-Scale Human Sketch Retrieval (2018, methodology): “Subsequent work further exploited stroke-level temporal information by applying heuristic data augmentation [33].”
Sketch-Guided Object Localization in Natural Images (2020, methodology): “Sketch Representation: Besides the traditionally used convolutional neural networks [39], there have been some recent advancements towards robust representation learning for sketches, e.”
Deep Learning for Free-Hand Sketch: A Survey (2020, background): “Given an ensemble of Sketch-a-Net trained by varying the mini-batch orders, joint Bayesian fusion [79] is adopted to improve the recognition performance.”
Cross-Modal Hierarchical Modelling for Fine-Grained Sketch Based Image Retrieval (2020, methodology): “2 Competitors We evaluate our model against three groups of competitors: (i) Evaluation against existing state-of-the-arts (SoA): Triplet-SN [50] uses Sketch-a-Net [51] as a baseline feature extractor trained using triplet ranking loss.”
Learning to Sketch with Shortcut Cycle Consistency (2018, neutral): “Previous work [47] resorts to heuristic stroke augmentation and removal techniques to enhance the training data.”
Semantically Tied Paired Cycle Consistency for Zero-Shot Sketch-Based Image Retrieval (2019, neutral): “SaN [78] 0.”
Vectorization and Rasterization: Self-Supervised Learning for Sketch and Handwriting (2021, methodology): “Sketch Analysis: We use sketch-recognition [65] and sketch-retrieval [62, 37] as downstream tasks to evaluate the quality of learned latent representation obtained by our self-supervised pre-training.”