Exploiting Pre-trained Feature Networks for Generative Adversarial Networks in Audio-domain Loop Generation

While generative adversarial networks (GANs) have been widely used in research on audio generation, the training of a GAN model is known to be unstable, time consuming, and data inefficient. Among the attempts to ameliorate the training process of GANs, the idea of Projected GAN emerges as an effective solution for GAN-based image generation, establishing the state-of-the-art in different image applications. The core idea is to use a pre-trained classifier to constrain the feature space of the discriminator to stabilize and improve GAN training. This paper investigates whether Projected GAN can similarly improve audio generation, by evaluating the performance of a StyleGAN2-based audio-domain loop generation model with and without using a pre-trained feature space in the discriminator. Moreover, we compare the performance of using a general versus domain-specific classifier as the pre-trained audio classifier. With experiments on both drum loop and synth loop generation, we show that a general audio classifier works better, and that with Projected GAN our loop generation models can converge around 5 times faster without performance degradation.

IronDepth: Iterative Refinement of Single-View Depth using Surface Normal and its Uncertainty

Single image surface normal estimation and depth estimation are closely related problems as the former can be calculated from the latter. However, the surface normals computed from the output of depth estimation methods are significantly less accurate than the surface normals directly estimated by networks. To reduce such discrepancy, we introduce a novel framework that uses surface normal and its uncertainty to recurrently refine the predicted depth-map. The depth of each pixel can be propagated to a query pixel, using the predicted surface normal as guidance. We thus formulate depth refinement as a classification of choosing the neighboring pixel to propagate from. Then, by propagating to sub-pixel points, we upsample the refined, low-resolution output. The proposed method shows state-of-the-art performance on NYUv2 and iBims-1 - both in terms of depth and normal. Our refinement module can also be attached to the existing depth estimation methods to improve their accuracy. We also show that our framework, only trained for depth estimation, can also be used for depth completion. The code is available at https://github.com/baegwangbin/IronDepth.

Atomized Deep Learning Models

Deep learning models often tackle the intra-sample structure, such as the order of words in a sentence and pixels in an image, but have not pay much attention to the inter-sample relationship. In this paper, we show that explicitly modeling the inter-sample structure to be more discretized can potentially help model's expressivity. We propose a novel method, Atom Modeling, that can discretize a continuous latent space by drawing an analogy between a data point and an atom, which is naturally spaced away from other atoms with distances depending on their intra structures. Specifically, we model each data point as an atom composed of electrons, protons, and neutrons and minimize the potential energy caused by the interatomic force among data points. Through experiments with qualitative analysis in our proposed Atom Modeling on synthetic and real datasets, we find that Atom Modeling can improve the performance by maintaining the inter-sample relation and can capture an interpretable intra-sample relation by mapping each component in a data point to electron/proton/neutron.

4D Unsupervised Object Discovery

Object discovery is a core task in computer vision. While fast progresses have been made in supervised object detection, its unsupervised counterpart remains largely unexplored. With the growth of data volume, the expensive cost of annotations is the major limitation hindering further study. Therefore, discovering objects without annotations has great significance. However, this task seems impractical on still-image or point cloud alone due to the lack of discriminative information. Previous studies underlook the crucial temporal information and constraints naturally behind multi-modal inputs. In this paper, we propose 4D unsupervised object discovery, jointly discovering objects from 4D data -- 3D point clouds and 2D RGB images with temporal information. We present the first practical approach for this task by proposing a ClusterNet on 3D point clouds, which is jointly iteratively optimized with a 2D localization network. Extensive experiments on the large-scale Waymo Open Dataset suggest that the localization network and ClusterNet achieve competitive performance on both class-agnostic 2D object detection and 3D instance segmentation, bridging the gap between unsupervised methods and full supervised ones. Codes and models will be made available at https://github.com/Robertwyq/LSMOL.

Underwater Ranker: Learn Which Is Better and How to Be Better

In this paper, we present a ranking-based underwater image quality assessment (UIQA) method, abbreviated as URanker. The URanker is built on the efficient conv-attentional image Transformer. In terms of underwater images, we specially devise (1) the histogram prior that embeds the color distribution of an underwater image as histogram token to attend global degradation and (2) the dynamic cross-scale correspondence to model local degradation. The final prediction depends on the class tokens from different scales, which comprehensively considers multi-scale dependencies. With the margin ranking loss, our URanker can accurately rank the order of underwater images of the same scene enhanced by different underwater image enhancement (UIE) algorithms according to their visual quality. To achieve that, we also contribute a dataset, URankerSet, containing sufficient results enhanced by different UIE algorithms and the corresponding perceptual rankings, to train our URanker. Apart from the good performance of URanker, we found that a simple U-shape UIE network can obtain promising performance when it is coupled with our pre-trained URanker as additional supervision. In addition, we also propose a normalization tail that can significantly improve the performance of UIE networks. Extensive experiments demonstrate the state-of-the-art performance of our method. The key designs of our method are discussed. We will release our dataset and code.

Understanding Pure CLIP Guidance for Voxel Grid NeRF Models

We explore the task of text to 3D object generation using CLIP. Specifically, we use CLIP for guidance without access to any datasets, a setting we refer to as pure CLIP guidance. While prior work has adopted this setting, there is no systematic study of mechanics for preventing adversarial generations within CLIP. We illustrate how different image-based augmentations prevent the adversarial generation problem, and how the generated results are impacted. We test different CLIP model architectures and show that ensembling different models for guidance can prevent adversarial generations within bigger models and generate sharper results. Furthermore, we implement an implicit voxel grid model to show how neural networks provide an additional layer of regularization, resulting in better geometrical structure and coherency of generated objects. Compared to prior work, we achieve more coherent results with higher memory efficiency and faster training speeds.

LAFITE: Towards Language-Free Training for Text-to-Image Generation

One of the major challenges in training text-to-image generation models is the need of a large number of high-quality image-text pairs. While image samples are often easily accessible, the associated text descriptions typically require careful human captioning, which is particularly time- and cost-consuming. In this paper, we propose the first work to train text-to-image generation models without any text data. Our method leverages the well-aligned multi-modal semantic space of the powerful pre-trained CLIP model: the requirement of text-conditioning is seamlessly alleviated via generating text features from image features. Extensive experiments are conducted to illustrate the effectiveness of the proposed method. We obtain state-of-the-art results in the standard text-to-image generation tasks. Importantly, the proposed language-free model outperforms most existing models trained with full image-text pairs. Furthermore, our method can be applied in fine-tuning pre-trained models, which saves both training time and cost in training text-to-image generation models. Our pre-trained model obtains competitive results in zero-shot text-to-image generation on the MS-COCO dataset, yet with around only 1% of the model size and training data size relative to the recently proposed large DALL-E model.

Custom Structure Preservation in Face Aging

In this work, we propose a novel architecture for face age editing that can produce structural modifications while maintaining relevant details present in the original image. We disentangle the style and content of the input image and propose a new decoder network that adopts a style-based strategy to combine the style and content representations of the input image while conditioning the output on the target age. We go beyond existing aging methods allowing users to adjust the degree of structure preservation in the input image during inference. To this purpose, we introduce a masking mechanism, the CUstom Structure Preservation module, that distinguishes relevant regions in the input image from those that should be discarded. CUSP requires no additional supervision. Finally, our quantitative and qualitative analysis which include a user study, show that our method outperforms prior art and demonstrates the effectiveness of our strategy regarding image editing and adjustable structure preservation. Code and pretrained models are available at https://github.com/guillermogotre/CUSP.

Target and Task specific Source-Free Domain Adaptive Image Segmentation

Solving the domain shift problem during inference is essential in medical imaging as most deep-learning based solutions suffer from it. In practice, domain shifts are tackled by performing Unsupervised Domain Adaptation (UDA), where a model is adapted to an unlabeled target domain by leveraging the labelled source domain. In medical scenarios, the data comes with huge privacy concerns making it difficult to apply standard UDA techniques. Hence, a closer clinical setting is Source-Free UDA (SFUDA), where we have access to source trained model but not the source data during adaptation. Methods trying to solve SFUDA typically address the domain shift using pseudo-label based self-training techniques. However, due to domain shift, these pseudo-labels are usually of high entropy and denoising them still does not make them perfect labels to supervise the model. Therefore, adapting the source model with noisy pseudo labels reduces its segmentation capability while addressing the domain shift. To this end, we propose a two-stage approach for source-free domain adaptive image segmentation: 1) Target-specific adaptation followed by 2) Task-specific adaptation. In the first stage, we focus on generating target-specific pseudo labels while suppressing high entropy regions by proposing an Ensemble Entropy Minimization loss. We also introduce a selective voting strategy to enhance pseudo-label generation. In the second stage, we focus on adapting the network for task-specific representation by using a teacher-student self-training approach based on augmentation-guided consistency. We evaluate our proposed method on both 2D fundus datasets and 3D MRI volumes across 7 different domain shifts where we achieve better performance than recent UDA and SF-UDA methods for medical image segmentation. Code is available at https://github.com/Vibashan/tt-sfuda.

Prototypical few-shot segmentation for cross-institution male pelvic structures with spatial registration

The prowess that makes few-shot learning desirable in medical image analysis is the efficient use of the support image data, which are labelled to classify or segment new classes, a task that otherwise requires substantially more training images and expert annotations. This work describes a fully 3D prototypical few-shot segmentation algorithm, such that the trained networks can be effectively adapted to clinically interesting structures that are absent in training, using only a few labelled images from a different institute. First, to compensate for the widely recognised spatial variability between institutions in episodic adaptation of novel classes, a novel spatial registration mechanism is integrated into prototypical learning, consisting of a segmentation head and an spatial alignment module. Second, to assist the training with observed imperfect alignment, support mask conditioning module is proposed to further utilise the annotation available from the support images. Extensive experiments are presented in an application of segmenting eight anatomical structures important for interventional planning, using a data set of 589 pelvic T2-weighted MR images, acquired at seven institutes. The results demonstrate the efficacy in each of the 3D formulation, the spatial registration, and the support mask conditioning, all of which made positive contributions independently or collectively. Compared with the previously proposed 2D alternatives, the few-shot segmentation performance was improved with statistical significance, regardless whether the support data come from the same or different institutes.