Pradyumna Reddy

I am currently a Senior Research Scientist at Universal Foundation Model Research Team Autodesk Research London. I was previously a Senior Research Scientist at Noah's Ark Lab Huawei Research London.

Before that I was a PhD student working with Smart geometry Group, at University College London. My PhD advisor is Prof. Niloy J. Mitra. During my PhD I have had the opportunity to intern at Google Brain Amsterdam(Now Deepmind) and 2 x Adobe Research Creative Intelligence Lab.

I completed my B.E Hons Computer Science and Engineering from BITS Pilani Goa, and did my undergrad-thesis at Harvard Medical School(Laboratory of Mathematics of Imaging, Psychiatry NeuroImaging Laboratory), where I worked with Prof. Yogesh Rathi. After that, I worked as a Statistical Analyst with the Data and Analytics group at Walmart Labs for a couple of years. Following which I joined UCL.

CV /  LinkedIn  /  Email  /  Google Scholar  /  GitHub

Research

Currently, my research is focused on high-resolution 3D generative models.

Over the years, I have worked on diverse topics that include tractography(Brain Imaging), vector graphics, graphs, and audio.

During my PhD I am broadly working on using learning-based algorithms for analysis of unstructured data like graphs, vector graphics, and 3D data representations.

Publications

Please find an updated list here: Google Scholar.

3D-WAG: Hierarchical Wavelet-Guided Autoregressive Generation for High-Fidelity 3D Shapes

Autoregressive (AR) models have achieved remarkable success in natural language and image generation, but their application to 3D shape modeling remains largely unexplored. Unlike diffusion models, AR models enable more efficient and controllable generation with faster inference times, making them especially suitable for data-intensive domains. Traditional 3D generative models using AR approaches often rely on next-token” predictions at the voxel or point level. While effective for certain applications, these methods can be restrictive and computationally expensive when dealing with large-scale 3D data. To tackle these challenges, we introduce 3D-WAG, an AR model for 3D implicit distance fields that can perform unconditional shape generation, class-conditioned and also text-conditioned shape generation. Our key idea is to encode shapes as multi-scale wavelet token maps and use a Transformer to predict the next higher-resolution token map” in an autoregressive manner. By redefining 3D AR generation task as next-scale prediction, we reduce the computational cost of generation compared to traditional next-token prediction models, while preserving essential geometric details of 3D shapes in a more structured and hierarchical manner. We evaluate 3D-WAG to showcase its benefit by quantitative and qualitative comparisons with state-of-the-art methods on widely used benchmarks. Our results show 3D-WAG achieves superior performance in key metrics like Coverage and MMD, generating high-fidelity 3D shapes that closely match the real data distribution.

Eq-WaLa: Equivariant Augmentation and Regularization for Wavelet Latent Flow Matching

Despite their strong performance in compact 3D shape modeling, latent implicit generative models typically neglect geometric symmetries of 3D shapes, particularly rotation invariance, causing representations to depend on arbitrary coordinate frames rather than intrinsic shape structure. To tackle this, we propose an equivariant generative framework for 3D shapes that explicitly accounts for rotational variability during training. Our method builds on a compact wavelet-tree representation of a shape’s signed distance field (SDF), enabling multi-scale, channel-efficient encoding and efficient convolutional processing. To address the common challenge of unaligned orientations in real-world datasets, we introduce two key components: (i) a novel wavelet-domain rotation augmentation scheme that enables exact on-the-fly construction of rotated inputs via the inherent equivariance of wavelet transforms, and (ii) a latent-space regularization strategy that enforces consistency under discrete latent rotations. Combined with a flow matching model trained on these augmented and regularized latents, our framework can produce high-quality 3D shapes. We validate the effectiveness of these components, showing improved reconstruction accuracy and faster generative convergence on the Thingi10K dataset, which lacks consistent canonical orientations. Furthermore, we also demonstrate the scalability of our approach to large-scale settings, including both unconditional and text-conditioned shape generation tasks on the diverse and complex Objaverse dataset.

Wavelet latent diffusion (wala): Billion-parameter 3d generative model with compact wavelet encodings

Large-scale 3D generative models require substantial computational resources yet often fall short in capturing fine details and complex geometries at high resolutions. We attribute this limitation to the inefficiency of current representations, which lack the compactness required to model the generative models effectively. To address this, we introduce a novel approach called Wavelet Latent Diffusion, or WaLa, that encodes 3D shapes into a wavelet-based, compact latent encodings. Specifically, we compress a 2563 signed distance field into a 123 × 4 latent grid, achieving an impressive 2,427× compression ratio with minimal loss of detail. This high level of compression allows our method to efficiently train large-scale generative networks without increasing the inference time. Our models, both conditional and unconditional, contain approximately one billion parameters and successfully generate high-quality 3D shapes at 2563 resolution. Moreover, WaLa offers rapid inference, producing shapes within two to four seconds depending on the condition, despite the model’s scale. We demonstrate state-of-the-art performance across multiple datasets, with significant improvements in generation quality, diversity, and computational efficiency. We open-source our code and, to the best of our knowledge, release the largest pretrained 3D generative models across different modalities.

Velora: Memory efficient training using rank-1 sub-token projections

Large language models (LLMs) have recently emerged as powerful tools for tackling many language-processing tasks. Despite their success, training and fine-tuning these models is still far too computationally and memory intensive. In this paper, we identify and characterise the important components needed for effective model convergence using gradient descent. In doing so we find that the intermediate activations used to implement backpropagation can be excessively compressed without incurring any degradation in performance. This result leads us to a cheap and memory-efficient algorithm for both fine-tuning and pre-training LLMs. The proposed algorithm simply divides the tokens up into smaller sub-tokens before projecting them onto a fixed 1-dimensional subspace during the forward pass. These features are then coarsely reconstructed during the backward pass to implement the update rules. We confirm the effectiveness of our algorithm as being complimentary to many state-of-the-art PEFT methods on the VTAB-1k fine-tuning benchmark. Furthermore, we outperform QLoRA for fine-tuning LLaMA and show competitive performance against other memory-efficient pre-training methods on the large-scale C4 dataset.

G3DR: Generative 3D Reconstruction in ImageNet

We introduce a novel 3D generative method, Generative 3D Reconstruction (G3DR) in ImageNet, capable of generating diverse and high-quality 3D objects from single images, addressing the limitations of existing methods. At the heart of our framework is a novel depth regularization technique that enables the generation of scenes with high-geometric fidelity. G3DR also leverages a pretrained language-vision model, such as CLIP, to enable reconstruction in novel views and improve the visual realism of generations. Additionally, G3DR designs a simple but effective sampling procedure to further improve the quality of generations. G3DR offers diverse and efficient 3D asset generation based on class or text conditioning. Despite its simplicity, G3DR is able to beat state-of-theart methods, improving over them by up to 22% in perceptual metrics and 90% in geometry scores, while needing only half of the training time.

AvatarMMC: 3D Head Avatar Generation and Editing with Multi-Modal Conditioning

We introduce an approach for 3D head avatar generation and editing with multi-modal conditioning based on a 3D Generative Adversarial Network (GAN) and a Latent Diffusion Model (LDM). 3D GANs can generate high-quality head avatars given a single or no condition. However, it is challenging to generate samples that adhere to multiple conditions of different modalities. On the other hand, LDMs excel at learning complex conditional distributions. To this end, we propose to exploit the conditioning capabilities of LDMs to enable multi-modal control over the latent space of a pre-trained 3D GAN. Our method can generate and edit 3D head avatars given a mixture of control signals such as RGB input, segmentation masks, and global attributes. This provides better control over the generation and editing of synthetic avatars both globally and locally. Experiments show that our proposed approach outperforms a solely GAN-based approach both qualitatively and quantitatively on generation and editing tasks. To the best of our knowledge, our approach is the first to introduce multi-modal conditioning to 3D avatar generation and editing.

AudioSlots: A slot-centric generative model for audio separation

In a range of recent works, object-centric architectures have been shown to be suitable for unsupervised scene decomposition in the vision domain. Inspired by these methods we present AudioSlots, a slot-centric generative model for blind source separation in the audio domain. AudioSlots is built using permutation-equivariant encoder and decoder networks. The encoder network based on the Transformer architecture learns to map a mixed audio spectrogram to an unordered set of independent source embeddings. The spatial broadcast decoder network learns to generate the source spectrograms from the source embeddings. We train the model in an end-to-end manner using a permutation invariant loss function. Our results on Libri2Mix speech separation constitute a proof of concept that this approach shows promise. We discuss the results and limitations of our approach in detail, and further outline potential ways to overcome the limitations and directions for future work.

Search for Concepts: Discovering Visual Concepts Using Direct Optimization

Finding an unsupervised decomposition of an image into individual objects is a key step to leverage compositionality and to perform symbolic reasoning. Traditionally, this problem is solved using amortized inference, which does not generalize beyond the scope of the training data, may sometimes miss correct decompositions, and requires large amounts of training data. We propose finding a decomposition using direct, unamortized optimization, via a combination of a gradient-based optimization for differentiable object properties and global search for non-differentiable properties. We show that using direct optimization is more generalizable, misses fewer correct decompositions, and typically requires less data than methods based on amortized inference. This highlights a weakness of the current prevalent practice of using amortized inference that can potentially be improved by integrating more direct optimization elements.

A Multi-Implicit Neural Representation for Fonts

Fonts are ubiquitous across documents and come in a variety of styles. They are either represented in a native vector format or rasterized to produce fixed resolution images. In the first case, the non-standard representation prevents benefiting from latest network architectures for neural representations; while, in the latter case, the rasterized representation, when encoded via networks, results in loss of data fidelity, as font-specific discontinuities like edges and corners are difficult to represent using neural networks. Based on the observation that complex fonts can be represented by a superposition of a set of simpler occupancy functions, we introduce \textit{multi-implicits} to represent fonts as a permutation-invariant set of learned implict functions, without losing features (e.g., edges and corners). However, while multi-implicits locally preserve font features, obtaining supervision in the form of ground truth multi-channel signals is a problem in itself. Instead, we propose how to train such a representation with only local supervision, while the proposed neural architecture directly finds globally consistent multi-implicits for font families. We extensively evaluate the proposed representation for various tasks including reconstruction, interpolation, and synthesis to demonstrate clear advantages with existing alternatives. Additionally, the representation naturally enables glyph completion, wherein a single characteristic font is used to synthesize a whole font family in the target style.

Im2Vec: Synthesizing Vector Graphics without Vector Supervision

Discovering Pattern Structure Using Differentiable Compositing

Patterns, which are collections of elements arranged in regular or near-regular arrangements, are an important graphic art form and widely used due to their elegant simplicity and aesthetic appeal. When a pattern is encoded as a flat image without the underlying structure, manually editing the pattern is tedious and challenging as one has to both preserve the individual element shapes and their original relative arrangements. State-of-the-art deep learning frameworks that operate at the pixel level are unsuitable for manipulating such patterns. Specifically, these methods can easily disturb the shapes of the individual elements or their arrangement, and thus fail to preserve the latent structures of the input patterns. We present a novel differentiable compositing operator using pattern elements and use it to discover structures, in the form of a layered representation of graphical objects, directly from raw pattern images. This operator allows us to adapt current deep learning based image methods to effectively handle patterns. We evaluate our method on a range of patterns and demonstrate superiority in the context of pattern manipulations when compared against state-of-the-art pixel-based pixel- or point-based alternatives.

SeeThrough: Finding Objects in Heavily Occluded Indoor Scene Images

Discovering 3D arrangements of objects from single indoor images is important given its many applications including interior design, content creation, etc. Although heavily researched in the recent years, existing approaches break down under medium or heavy occlusion as the core object detection module starts failing in absence of directly visible cues. Instead, we take into account holistic contextual 3D information, exploiting the fact that objects in indoor scenes co-occur mostly in typical near-regular configurations. First, we use a neural network trained on real indoor annotated images to extract 2D keypoints, and feed them to a 3D candidate object generation stage. Then, we solve a global selection problem among these 3D candidates using pairwise co-occurrence statistics discovered from a large 3D scene database. We iterate the process allowing for candidates with low keypoint response to be incrementally detected based on the location of the already discovered nearby objects. Focusing on chairs, we demonstrate significant performance improvement over combinations of state-of-the-art methods, especially for scenes with moderately to severely occluded objects.

Joint Multi-Fiber NODDI Parameter Estimation and Tractography Using the Unscented Information Filter

Tracing white matter fiber bundles is an integral part of analyzing brain connectivity. An accurate estimate of the underlying tissue parameters is also paramount in several neuroscience applications. In this work, we propose to use a joint fiber model estimation and tractography algorithm that uses the NODDI (neurite orientation dispersion diffusion imaging) model to estimate fiber orientation dispersion consistently and smoothly along the fiber tracts along with estimating the intracellular and extracellular volume fractions from the diffusion signal. While the NODDI model has been used in earlier works to estimate the microstructural parameters at each voxel independently, for the first time, we propose to integrate it into a tractography framework. We extend this framework to estimate the NODDI parameters for two crossing fibers, which is imperative to trace fiber bundles through crossings as well as to estimate the microstructural parameters for each fiber bundle separately. We propose to use the unscented information filter (UIF) to accurately estimate the model parameters and perform tractography. The proposed approach has significant computational performance improvements as well as numerical robustness over the unscented Kalman filter (UKF). Our method not only estimates the confidence in the estimated parameters via the covariance matrix, but also provides the Fisher-information matrix of the state variables (model parameters), which can be quite useful to measure model complexity. Results from in-vivo human brain data sets demonstrate the ability of our algorithm to trace through crossing fiber regions, while estimating orientation dispersion and other biophysical model parameters in a consistent manner along the tracts.

A Braille-based mobile communication and translation glove for deaf-blind people

Deafblind people are excluded from most forms of communication and information. This paper suggests a novel approach to support the communication and interaction of deaf- blind individuals, thus fostering their independence. It includes a smart glove that translates the Braille alphabet, which is used almost universally by the literate deafblind population, into text and vice versa, and communicates the message via SMS to a remote contact. It enables user to convey simple messages by capacitive touch sensors as input sensors placed on the palmer side of the glove and converted to text by the PC/mobile phone. The wearer can perceive and interpret incoming messages by tactile feedback patterns of mini vibrational motors on the dorsal side of the glove. The successful implementation of real-time two- way translation between English and Braille, and communication of the wearable device with a mobile phone/PC opens up new opportunities of information exchange which were hitherto un-available to deafblind individuals, such as remote communication, as well as parallel one-to many broadcast. The glove also makes communicating with laypersons without knowledge of Braille possible, without the need for trained interpreters.

Selective Visualization of Anomalies in Fundus Images via Sparse and Low Rank Decomposition

Design and source code from Jon Barron's website