Humans have a remarkable ability to disentangle complex sensory inputs (e.g., image, text) into simple factors of variation (e.g., shape, color) without much supervision. This ability has inspired many works that attempt to solve the following question: how do we invert the data generation process to extract those factors with minimal or no supervision? Several works in the literature on non-linear independent component analysis have established this negative result; without some knowledge of the data generation process or appropriate inductive biases, it is impossible to perform this inversion. In recent years, a lot of progress has been made on disentanglement under structural assumptions, e.g., when we have access to auxiliary information that makes the factors of variation conditionally independent. However, existing work requires a lot of auxiliary information, e.g., in supervised classification, it prescribes that the number of label classes should be at least equal to the total dimension of all factors of variation. In this work, we depart from these assumptions and ask: a) How can we get disentanglement when the auxiliary information does not provide conditional independence over the factors of variation? b) Can we reduce the amount of auxiliary information required for disentanglement? For a class of models where auxiliary information does not ensure conditional independence, we show theoretically and experimentally that disentanglement (to a large extent) is possible even when the auxiliary information dimension is much less than the dimension of the true latent representation.
Face inpainting aims to complete the corrupted regions of the face images, which requires coordination between the completed areas and the non-corrupted areas. Recently, memory-oriented methods illustrate great prospects in the generation related tasks by introducing an external memory module to improve image coordination. However, such methods still have limitations in restoring the consistency and continuity for specificfacial semantic parts. In this paper, we propose the coarse-to-fine Memory-Disentangled Refinement Networks (MDRNets) for coordinated face inpainting, in which two collaborative modules are integrated, Disentangled Memory Module (DMM) and Mask-Region Enhanced Module (MREM). Specifically, the DMM establishes a group of disentangled memory blocks to store the semantic-decoupled face representations, which could provide the most relevant information to refine the semantic-level coordination. The MREM involves a masked correlation mining mechanism to enhance the feature relationships into the corrupted regions, which could also make up for the correlation loss caused by memory disentanglement. Furthermore, to better improve the inter-coordination between the corrupted and non-corrupted regions and enhance the intra-coordination in corrupted regions, we design InCo2 Loss, a pair of similarity based losses to constrain the feature consistency. Eventually, extensive experiments conducted on CelebA-HQ and FFHQ datasets demonstrate the superiority of our MDRNets compared with previous State-Of-The-Art methods.
Most consumer-grade digital cameras can only capture a limited range of luminance in real-world scenes due to sensor constraints. Besides, noise and quantization errors are often introduced in the imaging process. In order to obtain high dynamic range (HDR) images with excellent visual quality, the most common solution is to combine multiple images with different exposures. However, it is not always feasible to obtain multiple images of the same scene and most HDR reconstruction methods ignore the noise and quantization loss. In this work, we propose a novel learning-based approach using a spatially dynamic encoder-decoder network, HDRUNet, to learn an end-to-end mapping for single image HDR reconstruction with denoising and dequantization. The network consists of a UNet-style base network to make full use of the hierarchical multi-scale information, a condition network to perform pattern-specific modulation and a weighting network for selectively retaining information. Moreover, we propose a Tanh_L1 loss function to balance the impact of over-exposed values and well-exposed values on the network learning. Our method achieves the state-of-the-art performance in quantitative comparisons and visual quality. The proposed HDRUNet model won the second place in the single frame track of NITRE2021 High Dynamic Range Challenge.
Automatic detection of cracks in concrete surfaces based on image processing is a clear trend in modern civil engineering applications. Most infrastructure is made of concrete and cracks reveal degradation of the structural integrity of the facilities, which can lead to extreme structural failures. There are many approaches to overcome the difficulties in image-based crack detection, ranging from the pre-processing of the input image to the proper adjustment of efficient classifiers, passing through the essential feature selection step. This paper is related to the process of constructing features from images to allow a classifier to find the boundaries between images with and without cracks. The most common approaches to feature extraction are the convolutional techniques to extract relevant positional information from images and the filters for edge detection or background removal. Here we apply hashing techniques for the first time used for features extraction in this problem. The study of the classification capacity of hashes is carried out by comparing 5 different hash algorithms, 2 of which are based on wavelets. The effect of applying the z-transform on the images before calculating the hashes was also studied, which totals the study of 10 new features for this problem. A comparative study of 17 different algorithms from the scikit-learn library was carried out. The results show that 9 of the 10 features are relevant to the problem, as well as that the accuracy of the classifiers varied between 0.697 for the Naive-Bayes Gaussian classifier and 0.99 for the Random Forest (RF) classifier. The feature extraction algorithm developed in this work and the RF classifier algorithm is suitable for embedded applications, for example in inspection drones, as long as they are highly accurate and computationally light, both in terms of memory and processing time.
Face sketch synthesis has been widely used in multi-media entertainment and law enforcement. Despite the recent developments in deep neural networks, accurate and realistic face sketch synthesis is still a challenging task due to the diversity and complexity of human faces. Current image-to-image translation-based face sketch synthesis frequently encounters over-fitting problems when it comes to small-scale datasets. To tackle this problem, we present an end-to-end Memory Oriented Style Transfer Network (MOST-Net) for face sketch synthesis which can produce high-fidelity sketches with limited data. Specifically, an external self-supervised dynamic memory module is introduced to capture the domain alignment knowledge in the long term. In this way, our proposed model could obtain the domain-transfer ability by establishing the durable relationship between faces and corresponding sketches on the feature level. Furthermore, we design a novel Memory Refinement Loss (MR Loss) for feature alignment in the memory module, which enhances the accuracy of memory slots in an unsupervised manner. Extensive experiments on the CUFS and the CUFSF datasets show that our MOST-Net achieves state-of-the-art performance, especially in terms of the Structural Similarity Index(SSIM).
Coronavirus 2019 has brought severe challenges to social stability and public health worldwide. One effective way of curbing the epidemic is to require people to wear masks in public places and monitor mask-wearing states by utilizing suitable automatic detectors. However, existing deep learning based models struggle to simultaneously achieve the requirements of both high precision and real-time performance. To solve this problem, we propose an improved lightweight face mask detector based on YOLOv5, which can achieve an excellent balance of precision and speed. Firstly, a novel backbone ShuffleCANet that combines ShuffleNetV2 network with Coordinate Attention mechanism is proposed as the backbone. Afterwards, an efficient path aggression network BiFPN is applied as the feature fusion neck. Furthermore, the localization loss is replaced with alpha-CIoU in model training phase to obtain higher-quality anchors. Some valuable strategies such as data augmentation, adaptive image scaling, and anchor cluster operation are also utilized. Experimental results on AIZOO face mask dataset show the superiority of the proposed model. Compared with the original YOLOv5, the proposed model increases the inference speed by 28.3% while still improving the precision by 0.58%. It achieves the best mean average precision of 95.2% compared with other seven existing models, which is 4.4% higher than the baseline.
In this paper, we aim to improve the data efficiency of image captioning. We propose VisualGPT, a data-efficient image captioning model that leverages the linguistic knowledge from a large pretrained language model (LM). A crucial challenge is to balance between the use of visual information in the image and prior linguistic knowledge acquired from pretraining.We designed a novel self-resurrecting encoder-decoder attention mechanism to quickly adapt the pretrained LM as the language decoder on a small amount of in-domain training data. The pro-posed self-resurrecting activation unit produces sparse activations but is not susceptible to zero gradients. When trained on 0.1%, 0.5% and 1% of MSCOCO and Conceptual Captions, the proposed model, VisualGPT, surpasses strong image captioning baselines. VisualGPT outperforms the best baseline model by up to 10.8% CIDEr on MS COCO and up to 5.4% CIDEr on Conceptual Captions.We also perform a series of ablation studies to quantify the utility of each system component. To the best of our knowledge, this is the first work that improves data efficiency of image captioning by utilizing LM pretrained on unimodal data. Our code is available at: https://github.com/Vision-CAIR/VisualGPT.
The most recent year has witnessed the success of applying the Vision Transformer (ViT) for image classification. However, there are still evidences indicating that ViT often suffers following two aspects, i) the high computation and the memory burden from applying the multiple Transformer layers for pre-training on a large-scale dataset, ii) the over-fitting when training on small datasets from scratch. To address these problems, a novel method, namely, Vision Reservoir computing (ViR), is proposed here for image classification, as a parallel to ViT. By splitting each image into a sequence of tokens with fixed length, the ViR constructs a pure reservoir with a nearly fully connected topology to replace the Transformer module in ViT. Two kinds of deep ViR models are subsequently proposed to enhance the network performance. Comparative experiments between the ViR and the ViT are carried out on several image classification benchmarks. Without any pre-training process, the ViR outperforms the ViT in terms of both model and computational complexity. Specifically, the number of parameters of the ViR is about 15% even 5% of the ViT, and the memory footprint is about 20% to 40% of the ViT. The superiority of the ViR performance is explained by Small-World characteristics, Lyapunov exponents, and memory capacity.
Plane Wave imaging enables many applications that require high frame rates, including localisation microscopy, shear wave elastography, and ultra-sensitive Doppler. To alleviate the degradation of image quality with respect to conventional focused acquisition, typically, multiple acquisitions from distinctly steered plane waves are coherently (i.e. after time-of-flight correction) compounded into a single image. This poses a trade-off between image quality and achievable frame-rate. To that end, we propose a new deep learning approach, derived by formulating plane wave compounding as a linear inverse problem, that attains high resolution, high-contrast images from just 3 plane wave transmissions. Our solution unfolds the iterations of a proximal gradient descent algorithm as a deep network, thereby directly exploiting the physics-based generative acquisition model into the neural network design. We train our network in a greedy manner, i.e. layer-by-layer, using a combination of pixel, temporal, and distribution (adversarial) losses to achieve both perceptual fidelity and data consistency. Through the strong model-based inductive bias, the proposed architecture outperforms several standard benchmark architectures in terms of image quality, with a low computational and memory footprint.
Face images are rich data items that are useful and can easily be collected in many applications, such as in 1-to-1 face verification tasks in the domain of security and surveillance systems. Multiple methods have been proposed to protect an individual's privacy by perturbing the images to remove traces of identifiable information, such as gender or race. However, significantly less attention has been given to the problem of protecting images while maintaining optimal task utility. In this paper, we study the novel problem of creating privacy-preserving image representations with respect to a given utility task by proposing a principled framework called the Adversarial Image Anonymizer (AIA). AIA first creates an image representation using a generative model, then enhances the learned image representations using adversarial learning to preserve privacy and utility for a given task. Experiments were conducted on a publicly available data set to demonstrate the effectiveness of AIA as a privacy-preserving mechanism for face images.