Name

This paper proposes a practical approach for the IPR protection on recurrent neural networks (RNN) with the Gatekeeper concept.

We propose novel passport-based DNN ownership verification schemes which are both robust to network modifications and resilient to ambiguity attacks. Extension of NeurIPS2019 (acceptance rate: 1428/6743 ~ 21.18%).

This paper proposed CyEDA to perform global level domain adaptation (DA) without any pre-trained networks integration/annotation labels.

We propose a reversed OFA Network training algorithm - Progressive Expansion (ProX) that achieve up to 68% training time reduction.

This work deals with the problem of low-light image enhancement, in the context of scene texts. Particularly, a novel image enhancement framework is proposed to precisely restore the scene texts, as well as the overall quality of the image simultaneously under extremely low-light images conditions.

This paper presents a complete IPR protection framework on multi-model DNN.

For the first time in image captioning research, we propose a novel end-to-end weight pruning method that performs gradual sparsification based on weight sensitivity to the training loss.

This paper proposes a novel deep hashing model with only a single learning objective which is a simplification from most state of the art papers generally use a large number (>4) of losses and regularizer. Further, with this learning objective, code balancing can be achieved by simply using a Batch Normalization (BN) layer and multi-label classification is also straightforward with label smoothing.

This paper presents a complete protection framework in both black-box and white-box settings to enforce IPR protection on GANs.

Oral cancer is a major health issue among low- and middleincome countries due to the late diagnosis. In this paper, we develop a novel deep learning framework named D’OraCa to classify oral lesions using photographic images, and the results are consistent with clinicians

This paper aims to cast the doubts towards the reliability of the DNN with solid evidence particularly in Federated Learning environment.

We proposed a novel learning to hash deep neural network, which uses a differentiable hinge-like loss.

We proposed a new formulation to the VAE that allows users to condition on the style of the generated music.

We proposed a new bilingual scene text + evidence VQA dataset named STE-VQA that is annotated with 15,056 English and 13,006 Chinese QA pairs. Within this, an evidence-based measure of an algorithm’s capacity to reason is also proposed that requires the VQA model to provide a bounding box of the predicted answer.

To facilitate a new text detection research, we introduce the Total-Text dataset that consists of 1555 images with more than 3 different text orientations, one of a kind. Extension of ICDAR 2017 (Oral, acceptance rate: 52/409 ~ 12.7%).

This paper leverages on Mask RCNN to detect the salient regions in pixel level in the object layer; while, in the background layer, a CNN model is used to encode the scene features.

This paper introduces a compact image captioning model, and shows competitive results on MS-COCO and InstaPIC-1.1M datasets despite having an embedding vocabularly size that is 39x-99x smaller.

This work aims at shifting the focus of low-light enhancement research towards a more functional direction, that is enhancement that prioritizes feature retrieval. For this reasonm, this paper modeled low-light image enhancement as a distribution of localized enhancement functions using Gaussian Process trained at runtime with reference data generated from a CNN. Also, two new evaluation metrics - local features matching and intensity histogram similarity are also proposed.

This paper proposes for the first time a phrase-based hierarchical Long Short-Term Memory (phi-LSTM) model to generate image description. It encodes sentence as a sequence of combination of phrases and words, instead of a sequence of words alone as in classical solutions. Extension of ACCV 2016: phi-LSTM (acceptance rate: 143/590 ~ 24.2%).

ExDark: Exclusively Dark (ExDark) dataset which to the best of our knowledge, is the largest collection of low-light images to-date with image class and object level annotations.

ArtGAN: A series of new approaches to improve Generative Adversarial Network (GAN) for conditional image synthesis. One of the key innovation of ArtGAN is that, the gradient of the loss function w.r.t. the label (randomly assigned to each generated image) is back-propagated from the categorical discriminator to the generator. Extension of ICIP 2017: ArtGAN.

In this work, we study video saliency detection using Partial Differential Equation (PDE). Experiments on various human-action datasets show that our video saliency detection model performs favourably against the conventional solutions.

Classification of plants based on a multi-organ approach is very challenging. In this work, we present two deep frameworks (i.e. HGO-CNN & Plant-StructNet) to classify plant. Experiment results show that our proposed network outperforms the state of the art on the PlantClef2015 benchmark. Extension of ICIP 2017: HGO-CNN.

Dense crowd density estimation is one of the fundamental tasks in crowd analysis. In this work, we study crowd counting using the concept and principles of granular computing. That is, we learn discriminative crowd features from granules, that conforms to the outline between crowd and background (i.e. non-crowd) regions, for density estimation

This paper is one of the early work that studies how deep learning can be used in plant classification. It is found that venations of different order have been chosen to uniquely represent each of the plant species. We show that these findings fit with the hierarchical botanical definitions of leaf characters. Extension of ICIP 2015: Deep-Plant.

This paper studies crowd segmentation using the concept and principles of granular computing.

This paper preliminary studies on artwork classification using AlexNet. Particularly, we present a study on large-scale classification of fine-art paintings. Our objectives is to answer the question - is that does a machine have or able to understand “imagination” in paintings?

This paper studies a classifer that is capable to recognize human action with only partial action sample. Best student paper nomination at FUZZ-IEEE 2015.

This paper studies a view-invariant human model based on Poisson and fuzzy qualitative reasoning. Experimental results showed the reliability and effectiveness of our proposed viewpoint estimation framework by using the challenging IXMAS human action dataset. Extension of FUZZ-IEEE 2013.

This paper surveys a crowd behavior analysis with 180 references. Particularly, this survey provides the readers with a review of the state-of-the-art methods in crowd behavior analysis from the physics and biologically inspired perspectives

This paper investigates orthogonal planar search for coronary artery centerline extraction from computed tomography angiography. Extension of MVA 2013.

This paper studies zero-shot learning based on topic model. Extension of ICIP 2013.

This paper studies on scene classification using BK sub-triangle product inference mechanism. Paricularly, we proposed an improved Bandler and Kohout's CK subproduct.

This paper studies a new multimodal human-robot interaction framework to support elderly people in their daily life.

First fuzzy review paper in human motion analysis with 252 references.

This paper studies a high reliability inference engine that is rule-free, tolerate to uncertainty with interval-valued fuzzy sets and consider features weight. Extension of FUZZ-IEEE 2013.

This paper proposes fuzzy qualitative rank classifier to handle natural scene classification. The proposed FQRC provides a ranking interpretation instead of binary decision. Extension of FUZZ-IEEE 2012.

One of the first work that exploring curve text detection. A new dataset (CUTE80) is proposed.

This paper studies abrupt motion tracking using PSO. Extension of MVA 2013: SwATrack.

This paper identifies and localizes salient regions in a crowd scene, by transforming low-level features extracted from the crowd motion field into a global similarity structure.

This paper studies discriminative RF codebook learning and generative classifier learning in image understanding.

This paper studies a hybrid model to detect human skin. A new skin detection dataset (Pratheepan) is proposed.

This paper studies human motion analysis with fuzzy qualitative reasoning. Extension of FUZZ-IEEE 2008.

This paper studies a fusion between Gaussian Mixture Model-based clustering (GMM)- Qualitative Normalised Templates (QNT) for human motion analysis. The proposed method has been applied to the recognition of eight types of human motions

This paper models human upper and lower limbs using qualitative normalised templates. Application in visual surveillance.