Prateek Singhal

Sheng Liu, Xu Zhang, Nitesh Sekhar et al. · amazon-science

Empirical studies suggest that machine learning models trained with empirical risk minimization (ERM) often rely on attributes that may be spuriously correlated with the class labels. Such models typically lead to poor performance during inference for data lacking such correlations. In this work, we explicitly consider a situation where potential spurious correlations are present in the majority of training data. In contrast with existing approaches, which use the ERM model outputs to detect the samples without spurious correlations and either heuristically upweight or upsample those samples, we propose the logit correction (LC) loss, a simple yet effective improvement on the softmax cross-entropy loss, to correct the sample logit. We demonstrate that minimizing the LC loss is equivalent to maximizing the group-balanced accuracy, so the proposed LC could mitigate the negative impacts of spurious correlations. Our extensive experimental results further reveal that the proposed LC loss outperforms state-of-the-art solutions on multiple popular benchmarks by a large margin, an average 5.5\% absolute improvement, without access to spurious attribute labels. LC is also competitive with oracle methods that make use of the attribute labels. Code is available at https://github.com/shengliu66/LC.

Moulik Choraria, Xinbo Wu, Sourya Basu et al. · amazon-science

General purpose Vision Language Models (VLMs) have received tremendous interest in recent years, owing to their ability to learn rich vision-language correlations as well as their broad zero-shot competencies. One immensely popular line of work utilizes frozen unimodal models, by bridging vision representations to language using a trainable module called the QFormer. However, this method relies heavily on large-scale multimodal pretraining with huge computational overheads. To that end, we propose a more efficient framework for QFormer-based vision-language alignment. Our key idea relies on the observation that QFormer latents correspond more strongly to the frozen LLM's intermediate latent space. Consequently, instead of using QFormer latents as inputs to the LLM, we alter the framework by using the latents to directly condition the LLM latent space for image-to-text generation. We demonstrate the effectiveness of our approach against existing baselines in improving the efficiency of vision-language pretraining.

Jianwei Feng, Prateek Singhal

Style transfer for human face has been widely researched in recent years. Majority of the existing approaches work in 2D image domain and have 3D inconsistency issue when applied on different viewpoints of the same face. In this paper, we tackle the problem of 3D face style transfer which aims at generating stylized novel views of a 3D human face with multi-view consistency. We propose to use a neural radiance field (NeRF) to represent 3D human face and combine it with 2D style transfer to stylize the 3D face. We find that directly training a NeRF on stylized images from 2D style transfer brings in 3D inconsistency issue and causes blurriness. On the other hand, training a NeRF jointly with 2D style transfer objectives shows poor convergence due to the identity and head pose gap between style image and content image. It also poses challenge in training time and memory due to the need of volume rendering for full image to apply style transfer loss functions. We therefore propose a hybrid framework of NeRF and mesh rasterization to combine the benefits of high fidelity geometry reconstruction of NeRF and fast rendering speed of mesh. Our framework consists of three stages: 1. Training a NeRF model on input face images to learn the 3D geometry; 2. Extracting a mesh from the trained NeRF model and optimizing it with style transfer objectives via differentiable rasterization; 3. Training a new color network in NeRF conditioned on a style embedding to enable arbitrary style transfer to the 3D face. Experiment results show that our approach generates high quality face style transfer with great 3D consistency, while also enabling a flexible style control.

Shristi Das Biswas, Matthew Shreve, Xuelu Li et al.

Recent advancements in language-guided diffusion models for image editing are often bottle-necked by cumbersome prompt engineering to precisely articulate desired changes. An intuitive alternative calls on guidance from in-the-wild image exemplars to help users bring their imagined edits to life. Contemporary exemplar-based editing methods shy away from leveraging the rich latent space learnt by pre-existing large text-to-image (TTI) models and fall back on training with curated objective functions to achieve the task. Though somewhat effective, this demands significant computational resources and lacks compatibility with diverse base models and arbitrary exemplar count. On further investigation, we also find that these techniques restrict user control to only applying uniform global changes over the entire edited region. In this paper, we introduce a novel framework for progressive exemplar-driven editing with off-the-shelf diffusion models, dubbed PIXELS, to enable customization by providing granular control over edits, allowing adjustments at the pixel or region level. Our method operates solely during inference to facilitate imitative editing, enabling users to draw inspiration from a dynamic number of reference images, or multimodal prompts, and progressively incorporate all the desired changes without retraining or fine-tuning existing TTI models. This capability of fine-grained control opens up a range of new possibilities, including selective modification of individual objects and specifying gradual spatial changes. We demonstrate that PIXELS delivers high-quality edits efficiently, leading to a notable improvement in quantitative metrics as well as human evaluation. By making high-quality image editing more accessible, PIXELS has the potential to enable professional-grade edits to a wider audience with the ease of using any open-source image generation model.

Moulik Choraria, Xinbo Wu, Akhil Bhimaraju et al. · amazon-science

The hyperscaling of data and parameter count in transformer models is yielding diminishing performance improvement, especially when weighed against training costs. Such plateauing underlines a growing need for more efficient finetuning and inference, without sacrificing performance. This is particularly pressing for multimodal learning, where the overhead of processing multimodal tokens alongside language data often limits the practical viability of these systems. In parallel, advances in representation learning and interpretability have deepened our understanding of how such models process and encode information. Notably, recent work has uncovered implicit cross-modal alignment in the deeper layers of large pretrained models. Interestingly, this aligns with our own observations that models naturally defer most cross-modal token interactions to deeper stages of computation. Building on this, we propose a simple modification. Instead of concatenation with the language prompt at the start, we insert multimodal tokens directly into the middle, allowing them to entirely bypass the early layers. Our results with diverse modalities: 1) LLaVA \& BLIP for vision, 2) LTU for audio, and 3) MoLCA for molecular data, indicate that our method reduces computational costs during both training and inference, while at the very least, preserving, if not surpassing the performance of existing baselines. Our work has important implications for scaling and composing pretrained models in a resource-efficient manner.

Prateek Singhal, Ruffin White, Henrik Christensen

We present an on-line 3D visual object tracking framework for monocular cameras by incorporating spatial knowledge and uncertainty from semantic mapping along with high frequency measurements from visual odometry. Using a combination of vision and odometry that are tightly integrated we can increase the overall performance of object based tracking for semantic mapping. We present a framework for integration of the two data-sources into a coherent framework through information based fusion/arbitration. We demonstrate the framework in the context of OmniMapper[1] and present results on 6 challenging sequences over multiple objects compared to data obtained from a motion capture systems. We are able to achieve a mean error of 0.23m for per frame tracking showing 9% relative error less than state of the art tracker.

N. Dinesh Reddy, Prateek Singhal, K. Madhava Krishna

While the literature has been fairly dense in the areas of scene understanding and semantic labeling there have been few works that make use of motion cues to embellish semantic performance and vice versa. In this paper, we address the problem of semantic motion segmentation, and show how semantic and motion priors augments performance. We pro- pose an algorithm that jointly infers the semantic class and motion labels of an object. Integrating semantic, geometric and optical ow based constraints into a dense CRF-model we infer both the object class as well as motion class, for each pixel. We found improvement in performance using a fully connected CRF as compared to a standard clique-based CRFs. For inference, we use a Mean Field approximation based algorithm. Our method outperforms recently pro- posed motion detection algorithms and also improves the semantic labeling compared to the state-of-the-art Automatic Labeling Environment algorithm on the challenging KITTI dataset especially for object classes such as pedestrians and cars that are critical to an outdoor robotic navigation scenario.

Prateek Singhal, Aditya Deshpande, N Dinesh Reddy et al.

Detecting multiple planes in images is a challenging problem, but one with many applications. Recent work such as J-Linkage and Ordered Residual Kernels have focussed on developing a domain independent approach to detect multiple structures. These multiple structure detection methods are then used for estimating multiple homographies given feature matches between two images. Features participating in the multiple homographies detected, provide us the multiple scene planes. We show that these methods provide locally optimal results and fail to merge detected planar patches to the true scene planes. These methods use only residues obtained on applying homography of one plane to another as cue for merging. In this paper, we develop additional cues such as local consistency of planes, local normals, texture etc. to perform better classification and merging . We formulate the classification as an MRF problem and use TRWS message passing algorithm to solve non metric energy terms and complex sparse graph structure. We show results on challenging dataset common in robotics navigation scenarios where our method shows accuracy of more than 85 percent on average while being close or same as the actual number of scene planes.