SceneDreamer: Unbounded 3D Scene Generation from 2D Image Collections

In this work, we present SceneDreamer, an unconditional generative model for unbounded 3D scenes, which synthesizes large-scale 3D landscapes from random noises. Our framework is learned from in-the-wild 2D image collections only, without any 3D annotations. At the core of SceneDreamer is a principled learning paradigm comprising 1) an efficient yet expressive 3D scene representation, 2) a generative scene parameterization, and 3) an effective renderer that can leverage the knowledge from 2D images. Our framework starts from an efficient bird's-eye-view (BEV) representation generated from simplex noise, which consists of a height field and a semantic field. The height field represents the surface elevation of 3D scenes, while the semantic field provides detailed scene semantics. This BEV scene representation enables 1) representing a 3D scene with quadratic complexity, 2) disentangled geometry and semantics, and 3) efficient training. Furthermore, we propose a novel generative neural hash grid to parameterize the latent space given 3D positions and the scene semantics, which aims to encode generalizable features across scenes. Lastly, a neural volumetric renderer, learned from 2D image collections through adversarial training, is employed to produce photorealistic images. Extensive experiments demonstrate the effectiveness of SceneDreamer and superiority over state-of-the-art methods in generating vivid yet diverse unbounded 3D worlds.

Vision Learners Meet Web Image-Text Pairs

Most recent self-supervised learning~(SSL) methods are pre-trained on the well-curated ImageNet-1K dataset. In this work, we consider SSL pre-training on noisy web image-text paired data due to the excellent scalability of web data. First, we conduct a benchmark study of representative SSL pre-training methods on large-scale web data in a fair condition. Methods include single-modal ones such as MAE and multi-modal ones such as CLIP. We observe that multi-modal methods cannot outperform single-modal ones on vision transfer learning tasks. We derive an information-theoretical view to explain the benchmarking results, which provides insights into designing novel vision learners. Inspired by the above explorations, we present a visual representation pre-training method, MUlti-modal Generator~(MUG), for scalable web image-text data. MUG achieves state-of-the-art transferring performances on a variety of tasks and shows promising scaling behavior. Models and codes will be made public. Demo available at https://huggingface.co/spaces/tennant/MUG_caption

FashionTex: Controllable Virtual Try-on with Text and Texture

Virtual try-on attracts increasing research attention as a promising way for enhancing the user experience for online cloth shopping. Though existing methods can generate impressive results, users need to provide a well-designed reference image containing the target fashion clothes that often do not exist. To support user-friendly fashion customization in full-body portraits, we propose a multi-modal interactive setting by combining the advantages of both text and texture for multi-level fashion manipulation. With the carefully designed fashion editing module and loss functions, FashionTex framework can semantically control cloth types and local texture patterns without annotated pairwise training data. We further introduce an ID recovery module to maintain the identity of input portrait. Extensive experiments have demonstrated the effectiveness of our proposed pipeline.

Discovering the Effectiveness of Pre-Training in a Large-scale Car-sharing Platform

Recent progress of deep learning has empowered various intelligent transportation applications, especially in car-sharing platforms. While the traditional operations of the car-sharing service highly relied on human engagements in fleet management, modern car-sharing platforms let users upload car images before and after their use to inspect the cars without a physical visit. To automate the aforementioned inspection task, prior approaches utilized deep neural networks. They commonly employed pre-training, a de-facto technique to establish an effective model under the limited number of labeled datasets. As candidate practitioners who deal with car images would presumably get suffered from the lack of a labeled dataset, we analyzed a sophisticated analogy into the effectiveness of pre-training is important. However, prior studies primarily shed a little spotlight on the effectiveness of pre-training. Motivated by the aforementioned lack of analysis, our study proposes a series of analyses to unveil the effectiveness of various pre-training methods in image recognition tasks at the car-sharing platform. We set two real-world image recognition tasks in the car-sharing platform in a live service, established them under the many-shot and few-shot problem settings, and scrutinized which pre-training method accomplishes the most effective performance in which setting. Furthermore, we analyzed how does the pre-training and fine-tuning convey different knowledge to the neural networks for a precise understanding.

Few-Shot Image-to-Semantics Translation for Policy Transfer in Reinforcement Learning

We investigate policy transfer using image-to-semantics translation to mitigate learning difficulties in vision-based robotics control agents. This problem assumes two environments: a simulator environment with semantics, that is, low-dimensional and essential information, as the state space, and a real-world environment with images as the state space. By learning mapping from images to semantics, we can transfer a policy, pre-trained in the simulator, to the real world, thereby eliminating real-world on-policy agent interactions to learn, which are costly and risky. In addition, using image-to-semantics mapping is advantageous in terms of the computational efficiency to train the policy and the interpretability of the obtained policy over other types of sim-to-real transfer strategies. To tackle the main difficulty in learning image-to-semantics mapping, namely the human annotation cost for producing a training dataset, we propose two techniques: pair augmentation with the transition function in the simulator environment and active learning. We observed a reduction in the annotation cost without a decline in the performance of the transfer, and the proposed approach outperformed the existing approach without annotation.

Image Labels Are All You Need for Coarse Seagrass Segmentation

Seagrass meadows serve as critical carbon sinks, but accurately estimating the amount of carbon they store requires knowledge of the seagrass species present. Using underwater and surface vehicles equipped with machine learning algorithms can help to accurately estimate the composition and extent of seagrass meadows at scale. However, previous approaches for seagrass detection and classification have required full supervision from patch-level labels. In this paper, we reframe seagrass classification as a weakly supervised coarse segmentation problem where image-level labels are used during training (25 times fewer labels compared to patch-level labeling) and patch-level outputs are obtained at inference time. To this end, we introduce SeaFeats, an architecture that uses unsupervised contrastive pretraining and feature similarity to separate background and seagrass patches, and SeaCLIP, a model that showcases the effectiveness of large language models as a supervisory signal in domain-specific applications. We demonstrate that an ensemble of SeaFeats and SeaCLIP leads to highly robust performance, with SeaCLIP conservatively predicting the background class to avoid false seagrass misclassifications in blurry or dark patches. Our method outperforms previous approaches that require patch-level labels on the multi-species 'DeepSeagrass' dataset by 6.8% (absolute) for the class-weighted F1 score, and by 12.1% (absolute) F1 score for seagrass presence/absence on the 'Global Wetlands' dataset. We also present two case studies for real-world deployment: outlier detection on the Global Wetlands dataset, and application of our method on imagery collected by FloatyBoat, an autonomous surface vehicle.

ID2image: Leakage of non-ID information into face descriptors and inversion from descriptors to images

Embedding a face image to a descriptor vector using a deep CNN is a widely used technique in face recognition. Via several possible training strategies, such embeddings are supposed to capture only identity information. Information about the environment (such as background and lighting) or changeable aspects of the face (such as pose, expression, presence of glasses, hat etc.) should be discarded since they are not useful for recognition. In this paper, we present a surprising result that this is not the case. We show that non-ID attributes, as well as landmark positions and the image histogram can be recovered from the ID embedding of state-of-the-art face embedding networks (VGGFace2 and ArcFace). In fact, these non-ID attributes can be predicted from ID embeddings with similar accuracy to a prediction from the original image. Going further, we present an optimisation strategy that uses a generative model (specifically StyleGAN2 for faces) to recover images from an ID embedding. We show photorealistic inversion from ID embedding to face image in which not only is the ID realistically reconstructed but the pose, lighting and background/apparel to some extent as well.

Deep Learning-Based UAV Aerial Triangulation without Image Control Points

The emerging drone aerial survey has the advantages of low cost, high efficiency, and flexible use. However, UAVs are often equipped with cheap POS systems and non-measurement cameras, and their flight attitudes are easily affected. How to realize the large-scale mapping of UAV image-free control supported by POS faces many technical problems. The most basic and important core technology is how to accurately realize the absolute orientation of images through advanced aerial triangulation technology. In traditional aerial triangulation, image matching algorithms are constrained to varying degrees by preset prior knowledge. In recent years, deep learning has developed rapidly in the field of photogrammetric computer vision. It has surpassed the performance of traditional handcrafted features in many aspects. It has shown stronger stability in image-based navigation and positioning tasks, especially it has better resistance to unfavorable factors such as blur, illumination changes, and geometric distortion. Based on the introduction of the key technologies of aerial triangulation without image control points, this paper proposes a new drone image registration method based on deep learning image features to solve the problem of high mismatch rate in traditional methods. It adopts SuperPoint as the feature detector, uses the superior generalization performance of CNN to extract precise feature points from the UAV image, thereby achieving high-precision aerial triangulation. Experimental results show that under the same pre-processing and post-processing conditions, compared with the traditional method based on the SIFT algorithm, this method achieves suitable precision more efficiently, which can meet the requirements of UAV aerial triangulation without image control points in large-scale surveys.

LDMIC: Learning-based Distributed Multi-view Image Coding

Multi-view image compression plays a critical role in 3D-related applications. Existing methods adopt a predictive coding architecture, which requires joint encoding to compress the corresponding disparity as well as residual information. This demands collaboration among cameras and enforces the epipolar geometric constraint between different views, which makes it challenging to deploy these methods in distributed camera systems with randomly overlapping fields of view. Meanwhile, distributed source coding theory indicates that efficient data compression of correlated sources can be achieved by independent encoding and joint decoding, which motivates us to design a learning-based distributed multi-view image coding (LDMIC) framework. With independent encoders, LDMIC introduces a simple yet effective joint context transfer module based on the cross-attention mechanism at the decoder to effectively capture the global inter-view correlations, which is insensitive to the geometric relationships between images. Experimental results show that LDMIC significantly outperforms both traditional and learning-based MIC methods while enjoying fast encoding speed. Code will be released at https://github.com/Xinjie-Q/LDMIC.

SUG: Single-dataset Unified Generalization for 3D Point Cloud Classification

Although Domain Generalization (DG) problem has been fast-growing in the 2D image tasks, its exploration on 3D point cloud data is still insufficient and challenged by more complex and uncertain cross-domain variances with uneven inter-class modality distribution. In this paper, different from previous 2D DG works, we focus on the 3D DG problem and propose a Single-dataset Unified Generalization (SUG) framework that only leverages a single source dataset to alleviate the unforeseen domain differences faced by a well-trained source model. Specifically, we first design a Multi-grained Sub-domain Alignment (MSA) method, which can constrain the learned representations to be domain-agnostic and discriminative, by performing a multi-grained feature alignment process between the splitted sub-domains from the single source dataset. Then, a Sample-level Domain-aware Attention (SDA) strategy is presented, which can selectively enhance easy-to-adapt samples from different sub-domains according to the sample-level inter-domain distance to avoid the negative transfer. Experiments demonstrate that our SUG can boost the generalization ability for unseen target domains, even outperforming the existing unsupervised domain adaptation methods that have to access extensive target domain data. Our code is available at https://github.com/SiyuanHuang95/SUG.