Neural-network-based single image depth prediction (SIDP) is a challenging task where the goal is to predict the scene's per-pixel depth at test time. Since the problem, by definition, is ill-posed, the fundamental goal is to come up with an approach that can reliably model the scene depth from a set of training examples. In the pursuit of perfect depth estimation, most existing state-of-the-art learning techniques predict a single scalar depth value per-pixel. Yet, it is well-known that the trained model has accuracy limits and can predict imprecise depth. Therefore, an SIDP approach must be mindful of the expected depth variations in the model's prediction at test time. Accordingly, we introduce an approach that performs continuous modeling of per-pixel depth, where we can predict and reason about the per-pixel depth and its distribution. To this end, we model per-pixel scene depth using a multivariate Gaussian distribution. Moreover, contrary to the existing uncertainty modeling methods -- in the same spirit, where per-pixel depth is assumed to be independent, we introduce per-pixel covariance modeling that encodes its depth dependency w.r.t all the scene points. Unfortunately, per-pixel depth covariance modeling leads to a computationally expensive continuous loss function, which we solve efficiently using the learned low-rank approximation of the overall covariance matrix. Notably, when tested on benchmark datasets such as KITTI, NYU, and SUN-RGB-D, the SIDP model obtained by optimizing our loss function shows state-of-the-art results. Our method's accuracy (named MG) is among the top on the KITTI depth-prediction benchmark leaderboard.
Masked Autoencoders learn strong visual representations and achieve state-of-the-art results in several independent modalities, yet very few works have addressed their capabilities in multi-modality settings. In this work, we focus on point cloud and RGB image data, two modalities that are often presented together in the real world, and explore their meaningful interactions. To improve upon the cross-modal synergy in existing works, we propose PiMAE, a self-supervised pre-training framework that promotes 3D and 2D interaction through three aspects. Specifically, we first notice the importance of masking strategies between the two sources and utilize a projection module to complementarily align the mask and visible tokens of the two modalities. Then, we utilize a well-crafted two-branch MAE pipeline with a novel shared decoder to promote cross-modality interaction in the mask tokens. Finally, we design a unique cross-modal reconstruction module to enhance representation learning for both modalities. Through extensive experiments performed on large-scale RGB-D scene understanding benchmarks (SUN RGB-D and ScannetV2), we discover it is nontrivial to interactively learn point-image features, where we greatly improve multiple 3D detectors, 2D detectors, and few-shot classifiers by 2.9%, 6.7%, and 2.4%, respectively. Code is available at https://github.com/BLVLab/PiMAE.
Although numerous solutions have been proposed for image super-resolution, they are usually incompatible with low-power devices with many computational and memory constraints. In this paper, we address this problem by proposing a simple yet effective deep network to solve image super-resolution efficiently. In detail, we develop a spatially-adaptive feature modulation (SAFM) mechanism upon a vision transformer (ViT)-like block. Within it, we first apply the SAFM block over input features to dynamically select representative feature representations. As the SAFM block processes the input features from a long-range perspective, we further introduce a convolutional channel mixer (CCM) to simultaneously extract local contextual information and perform channel mixing. Extensive experimental results show that the proposed method is $3\times$ smaller than state-of-the-art efficient SR methods, e.g., IMDN, in terms of the network parameters and requires less computational cost while achieving comparable performance. The code is available at https://github.com/sunny2109/SAFMN.
Ultra-High-Definition (UHD) photo has gradually become the standard configuration in advanced imaging devices. The new standard unveils many issues in existing approaches for low-light image enhancement (LLIE), especially in dealing with the intricate issue of joint luminance enhancement and noise removal while remaining efficient. Unlike existing methods that address the problem in the spatial domain, we propose a new solution, UHDFour, that embeds Fourier transform into a cascaded network. Our approach is motivated by a few unique characteristics in the Fourier domain: 1) most luminance information concentrates on amplitudes while noise is closely related to phases, and 2) a high-resolution image and its low-resolution version share similar amplitude patterns.Through embedding Fourier into our network, the amplitude and phase of a low-light image are separately processed to avoid amplifying noise when enhancing luminance. Besides, UHDFour is scalable to UHD images by implementing amplitude and phase enhancement under the low-resolution regime and then adjusting the high-resolution scale with few computations. We also contribute the first real UHD LLIE dataset, \textbf{UHD-LL}, that contains 2,150 low-noise/normal-clear 4K image pairs with diverse darkness and noise levels captured in different scenarios. With this dataset, we systematically analyze the performance of existing LLIE methods for processing UHD images and demonstrate the advantage of our solution. We believe our new framework, coupled with the dataset, would push the frontier of LLIE towards UHD. The code and dataset are available at https://li-chongyi.github.io/UHDFour.
Previous works have shown that increasing the window size for Transformer-based image super-resolution models (e.g., SwinIR) can significantly improve the model performance but the computation overhead is also considerable. In this paper, we present SRFormer, a simple but novel method that can enjoy the benefit of large window self-attention but introduces even less computational burden. The core of our SRFormer is the permuted self-attention (PSA), which strikes an appropriate balance between the channel and spatial information for self-attention. Our PSA is simple and can be easily applied to existing super-resolution networks based on window self-attention. Without any bells and whistles, we show that our SRFormer achieves a 33.86dB PSNR score on the Urban100 dataset, which is 0.46dB higher than that of SwinIR but uses fewer parameters and computations. We hope our simple and effective approach can serve as a useful tool for future research in super-resolution model design.
We propose an image synthesis mechanism for multi-sequence prostate MR images conditioned on text, to control lesion presence and sequence, as well as to generate paired bi-parametric images conditioned on images e.g. for generating diffusion-weighted MR from T2-weighted MR for paired data, which are two challenging tasks in pathological image synthesis. Our proposed mechanism utilises and builds upon the recent stable diffusion model by proposing image-based conditioning for paired data generation. We validate our method using 2D image slices from real suspected prostate cancer patients. The realism of the synthesised images is validated by means of a blind expert evaluation for identifying real versus fake images, where a radiologist with 4 years experience reading urological MR only achieves 59.4% accuracy across all tested sequences (where chance is 50%). For the first time, we evaluate the realism of the generated pathology by blind expert identification of the presence of suspected lesions, where we find that the clinician performs similarly for both real and synthesised images, with a 2.9 percentage point difference in lesion identification accuracy between real and synthesised images, demonstrating the potentials in radiological training purposes. Furthermore, we also show that a machine learning model, trained for lesion identification, shows better performance (76.2% vs 70.4%, statistically significant improvement) when trained with real data augmented by synthesised data as opposed to training with only real images, demonstrating usefulness for model training.
Text-to-image synthesis refers to generating visual-realistic and semantically consistent images from given textual descriptions. Previous approaches generate an initial low-resolution image and then refine it to be high-resolution. Despite the remarkable progress, these methods are limited in fully utilizing the given texts and could generate text-mismatched images, especially when the text description is complex. We propose a novel Fine-grained text-image Fusion based Generative Adversarial Networks, dubbed FF-GAN, which consists of two modules: Fine-grained text-image Fusion Block (FF-Block) and Global Semantic Refinement (GSR). The proposed FF-Block integrates an attention block and several convolution layers to effectively fuse the fine-grained word-context features into the corresponding visual features, in which the text information is fully used to refine the initial image with more details. And the GSR is proposed to improve the global semantic consistency between linguistic and visual features during the refinement process. Extensive experiments on CUB-200 and COCO datasets demonstrate the superiority of FF-GAN over other state-of-the-art approaches in generating images with semantic consistency to the given texts.Code is available at https://github.com/haoranhfut/FF-GAN.
X-ray imaging is widely used for non-destructive detection of defects in industrial products on a conveyor belt. Real-time detection requires highly accurate, robust, and fast algorithms to analyze X-ray images. Deep convolutional neural networks (DCNNs) satisfy these requirements if a large amount of labeled data is available. To overcome the challenge of collecting these data, different methods of X-ray image generation can be considered. Depending on the desired level of similarity to real data, various physical effects either should be simulated or can be ignored. X-ray scattering is known to be computationally expensive to simulate, and this effect can heavily influence the accuracy of a generated X-ray image. We propose a methodology for quantitative evaluation of the effect of scattering on defect detection. This methodology compares the accuracy of DCNNs trained on different versions of the same data that include and exclude the scattering signal. We use the Probability of Detection (POD) curves to find the size of the smallest defect that can be detected with a DCNN and evaluate how this size is affected by the choice of training data. We apply the proposed methodology to a model problem of defect detection in cylinders. Our results show that the exclusion of the scattering signal from the training data has the largest effect on the smallest detectable defects. Furthermore, we demonstrate that accurate inspection is more reliant on high-quality training data for images with a high quantity of scattering. We discuss how the presented methodology can be used for other tasks and objects.
We propose SING (StabIlized and Normalized Gradient), a plug-and-play technique that improves the stability and generalization of the Adam(W) optimizer. SING is straightforward to implement and has minimal computational overhead, requiring only a layer-wise standardization of the gradients fed to Adam(W) without introducing additional hyper-parameters. We support the effectiveness and practicality of the proposed approach by showing improved results on a wide range of architectures, problems (such as image classification, depth estimation, and natural language processing), and in combination with other optimizers. We provide a theoretical analysis of the convergence of the method, and we show that by virtue of the standardization, SING can escape local minima narrower than a threshold that is inversely proportional to the network's depth.
Generative Adversarial Networks (GANs) have shown immense potential in fields far from physics, such as in text and image generation. Here we use GANs to learn a prototypical stochastic process on a lattice. By suitably adding noise to the original data we succeed in bringing both the Generator and the Discriminator loss functions close to their ideal value. However, as typical for adversarial approaches, oscillations persist. This undermines model selection and the quality of the generated trajectory. We demonstrate that a suitable multi-model procedure where stochastic trajectories are advanced at each step upon randomly selecting a Generator leads to a remarkable increase in accuracy. Based on the reported findings GANs appears as a promising tool to tackle complex statistical dynamics.