Animal3D consists of 3379 images collected from 40 mammal species, high-quality annotations of 26 keypoints, and importantly the pose and shape parameters of the SMAL model. We demonstrate that synthetic pre-training is a viable strategy to boost the model performance.

Diffusion models have emerged as a powerful method of generative modeling across a range of fields, capable of producing stunning photo-realistic images from natural language descriptions. However, these models lack explicit control over the 3D structure of the objects in the generated images. In this paper, we propose a novel method that incorporates 3D geometry control into diffusion models, making them generate even more realistic and diverse images. To achieve this, our method exploits ControlNet, which extends diffusion models by using visual prompts in addition to text prompts. We generate images of 3D objects taken from a 3D shape repository (e.g., ShapeNet and Objaverse), render them from a variety of poses and viewing directions, compute the edge maps of the rendered images, and use these edge maps as visual prompts to generate realistic images. With explicit 3D geometry control, we can easily change the 3D structures of the objects in the generated images and obtain ground-truth 3D annotations automatically. This allows us to use the generated images to improve a lot of vision tasks, e.g., classification and 3D pose estimation, in both in-distribution (ID) and out-of-distribution (OOD) settings.

Obtaining accurate 3D object poses is vital for numerous computer vision applications, such as 3D reconstruction and scene understanding. However, annotating real-world objects is time-consuming and challenging. While synthetically generated training data is a viable alternative, the domain shift between real and synthetic data is a significant challenge. In this work, we aim to narrow the performance gap between models trained on synthetic data and few real images and fully supervised models trained on large-scale data. We achieve this by approaching the problem from two perspectives: 1) We introduce SyntheticP3D, a new synthetic dataset for object pose estimation generated from CAD models and enhanced with a novel algorithm. 2) We propose a novel approach (CC3D) for training neural mesh models that perform pose estimation via inverse rendering. In particular, we exploit the spatial relationships between features on the mesh surface and a contrastive learning scheme to guide the domain adaptation process. Combined, these two approaches enable our models to perform competitively with state-of-the-art models using only 10% of the respective real training images, while outperforming the SOTA model by 10.4% with a threshold of pi/18 using only 50% of the real training data. Our trained model further demonstrates robust generalization to out-of-distribution scenarios despite being trained with minimal real data.

Obtaining accurate 3D object poses is vital for numerous computer vision applications, such as 3D reconstruction and scene understanding. However, annotating real-world objects is time-consuming and challenging. While synthetically generated training data is a viable alternative, the domain shift between real and synthetic data is a significant challenge. In this work, we aim to narrow the performance gap between models trained on synthetic data and few real images and fully supervised models trained on large-scale data. We achieve this by approaching the problem from two perspectives: 1) We introduce SyntheticP3D, a new synthetic dataset for object pose estimation generated from CAD models and enhanced with a novel algorithm. 2) We propose a novel approach (CC3D) for training neural mesh models that perform pose estimation via inverse rendering. In particular, we exploit the spatial relationships between features on the mesh surface and a contrastive learning scheme to guide the domain adaptation process. Combined, these two approaches enable our models to perform competitively with state-of-the-art models using only 10% of the respective real training images, while outperforming the SOTA model by 10.4% with a threshold of pi/18 using only 50% of the real training data. Our trained model further demonstrates robust generalization to out-of-distribution scenarios despite being trained with minimal real data.

Visual Question Answering (VQA) models often perform poorly on out-of-distribution data and struggle on domain generalization. Due to the multi-modal nature of this task, multiple factors of variation are intertwined, making generalization difficult to analyze. This motivates us to introduce a virtual benchmark, Super-CLEVR, where different factors in VQA domain shifts can be isolated in order that their effects can be studied independently. Four factors are considered: visual complexity, question redundancy, concept distribution and concept compositionality. With controllably generated data, Super-CLEVR enables us to test VQA methods in situations where the test data differs from the training data along each of these axes. We study four existing methods, including two neural symbolic methods NSCL and NSVQA, and two non-symbolic methods FiLM and mDETR; and our proposed method, probabilistic NSVQA (P-NSVQA), which extends NSVQA with uncertainty reasoning. P-NSVQA outperforms other methods on three of the four domain shift factors. Our results suggest that disentangling reasoning and perception, combined with probabilistic uncertainty, form a strong VQA model that is more robust to domain shifts

We consider the problem of category-level 6D pose estimation from a single RGB image. Our approach represents an object category as a cuboid mesh and learns a generative model of the neural feature activations at each mesh vertex to perform pose estimation through differentiable rendering. A common problem of rendering-based approaches is that they rely on bounding box proposals, which do not convey information about the 3D rotation of the object and are not reliable when objects are partially occluded. Instead, we introduce a coarse-to-fine optimization strategy that utilizes the rendering process to estimate a sparse set of 6D object proposals, which are subsequently refined with gradient-based optimization.

In this work, we introduce ROBIN, a benchmark dataset for diagnosing the robustness of vision algorithms to individual nuisances in real-world images. We provide results for a number of popular baselines and make several interesting observations. We believe our dataset provides a rich testbed to study the OOD robustness of vision algorithms and will help to significantly push forward research in this area.

Image/sketch completion is a core task that addresses the problem of completing the missing regions of an image/sketch with realistic and semantically consistent content. We address one type of completion which is producing a tentative completion of an aerial view of the remnants of a building structure. The inference process may start with as little as 10% of the structure and thus is fundamentally pluralistic (e.g., multiple completions are possible). We present a novel pluralistic building contour completion framework. A feature suggestion component uses an entropy-based model to request information from the user for the next most informative location in the image. Then, an image completion component trained using self-supervision and procedurally generated content produces a partial or full completion. In our synthetic and real-world experiments for archaeological sites in Turkey, with up to only 4 iterations, we complete building footprints having only 10–15% of the ancient structure initially visible. We also compare to various state-of-the-art methods and show our superior quantitative/qualitative performance. While we show results for archaeology, we anticipate our method can be used for restoring highly incomplete historical sketches and for modern day urban reconstruction despite occlusions.

Research project at Microsoft Research Asia supervised by Dr. Bin Li and Dr. Jiahao Li.

We propose a framework for making group decisions from natural language-based preferences. Experiments on the real world data confirms the efficacy of our method.

Characterize 10 different microstructure representations with image texture features and quantitative metrics from image segmentation. For the microstructure generation task, two schemes are considered: 1) generating high-resolution (1024x1024) microstructure images from random noise; and 2) train a style transfer GAN for image generation conditioned on the segmentation label.

Kinetic modeling of a discontinuous precipitation reaction (5 phases) by 1) deep learning with CNN, and 2) performing image segmentation of various microstructure and quantizing the area fractions.

We first review the application of image-driven machine learning approaches to the field of materials characterization. Then we analyze and discuss the impact of various approaches at each step of the experiments.