We propose a set of modifications that improve training time and likelihood estimation of hierarchical mixture models implemented via Probabilistic Circuits (PCs). Our proposal reduces the complexity of mutual information estimation in the structure learning step of PCs from quadratic to linear in the number of inputs, without sacrificing likelihood estimation performance on image datasets. We repurpose invertible transformations from the lossless compression community to improve likelihood estimation by a factor of up to 25% on benchmark image datasets, making PCs competitive with current standard codecs on low-resolution datasets. Despite our improvements, experiments with low- and high-resolution image datasets indicate that the advantage of lossless neural compression and PCs over standard codecs, such as WebP, disappears as the image size increases, motivating future work on practical lossless neural compression.

Data compression is a critical component of today's digital infrastructure, in particular for storage and transmission of visual data over the internet. While end-to-end neural codecs have been shown to outperform hand-designed image and video codecs, their mainstream adoption is hindered by their high computational cost. In this paper we reassess standard design choices for neural image compression autoencoders, identifying two inefficiencies: the first is the substantial compute spent on high-resolution feature maps, and the second is methods of applying activations that lead to less expressive representations. We mitigate both issues and propose a new architecture, PatchMixer, that begins with approximately patchwise-independent encoding, followed by mixing layers, thus enabling compute savings. PatchMixer achieves a BD-rate within 5% of VVC on Kodak with 97 kMACs per pixel of decoding complexity and 200 kMACs per pixel of cumulative complexity, to our knowledge the lowest cumulative complexity for a neural codec within 5% of VVC.

Recent advances in text-to-image (T2I) models have achieved impressive quality and consistency. However, this has come at the cost of representation diversity. While automatic evaluation methods exist for benchmarking model diversity, they either require reference image datasets or lack specificity about the kind of diversity measured, limiting their adaptability and interpretability. To address this gap, we introduce the Does-it/Can-it framework, DIM-CIM, a reference-free measurement of default-mode diversity ("Does" the model generate images with expected attributes?) and generalization capacity ("Can" the model generate diverse attributes for a particular concept?). We construct the COCO-DIMCIM benchmark, which is seeded with COCO concepts and captions and augmented by a large language model. With COCO-DIMCIM, we find that widely-used models improve in generalization at the cost of default-mode diversity when scaling from 1.5B to 8.1B parameters. DIMCIM also identifies fine-grained failure cases, such as attributes that are generated with generic prompts but are rarely generated when explicitly requested. Finally, we use DIMCIM to evaluate the training data of a T2I model and observe a correlation of 0.85 between diversity in training images and default-mode diversity. Our work provides a flexible and interpretable framework for assessing T2I model diversity and generalization, enabling a more comprehensive understanding of model performance.

Tokenization is associated with many poorly understood shortcomings in language models (LMs), yet remains an important component for long sequence scaling purposes. This work studies how tokenization impacts model performance by analyzing and comparing the stochastic behavior of tokenized models with their byte-level, or token-free, counterparts. We discover that, even when the two models are statistically equivalent, their predictive distributions over the next byte can be substantially different, a phenomenon we term as ``tokenization bias''. To fully characterize this phenomenon, we introduce the Byte-Token Representation Lemma, a framework that establishes a mapping between the learned token distribution and its equivalent byte-level distribution. From this result, we develop a next-byte sampling algorithm that eliminates tokenization bias without requiring further training or optimization. In other words, this enables zero-shot conversion of tokenized LMs into statistically equivalent token-free ones. We demonstrate its broad applicability with two use cases: fill-in-the-middle (FIM) tasks and model ensembles. In FIM tasks where input prompts may terminate mid-token, leading to out-of-distribution tokenization, our method mitigates performance degradation and achieves 18\% improvement in FIM coding benchmarks, while consistently outperforming the standard token healing fix. For model ensembles where each model employs a distinct vocabulary, our approach enables seamless integration, resulting in improved performance up to 3.7\% over individual models across various standard baselines in reasoning, knowledge, and coding. Code is available at:https: //github.com/facebookresearch/Exact-Byte-Level-Probabilities-from-Tokenized-LMs.

Vector quantization is a fundamental technique for compression and large-scale nearest neighbor search. For high-accuracy operating points, multi-codebook quantization associates data vectors with one element from each of multiple codebooks. An example is residual quantization (RQ), which iteratively quantizes the residual error of previous steps. Dependencies between the different parts of the code are, however, ignored in RQ, which leads to suboptimal rate-distortion performance. Qinco recently addressed this inefficiency by using a neural network to determine the quantization codebook in RQ based on the vector reconstruction from previous steps. In this paper we introduce Qinco2 which extends and improves Qinco with (i) improved vector encoding using codeword pre-selection and beam-search, (ii) a fast approximate decoder leveraging codeword pairs to establish accurate short-lists for search, and (iii) an optimized training procedure and network architecture. We conduct experiments on four datasets to evaluate Qinco2 for vector compression and billion-scale nearest neighbor search. We obtain outstanding results in both settings, improving the state-of-the-art reconstruction MSE by 44% for 16-byte vector compression on BigANN, and search accuracy by 24% with 8-byte encodings on Deep1M.

Image codecs are typically optimized to trade-off bitrate vs. distortion metrics. At low bitrates, this leads to compression artefacts which are easily perceptible, even when training with perceptual or adversarial losses. To improve image quality and remove dependency on the bitrate we propose to decode with iterative diffusion models. We condition the decoding process on a vector-quantized image representation, as well as a global image description to provide additional context. We dub our model `PerCo'' for ``perceptual compression'', and compare it to state-of-the-art codecs at rates from 0.1 down to 0.003 bits per pixel. The latter rate is more than an order of magnitude smaller than those considered in most prior work, compressing a 512x768 Kodak image with less than 153 bytes. Despite this ultra-low bitrate, our approach maintains the ability to reconstruct realistic images. We find that our model leads to reconstructions with state-of-the-art visual quality as measured by FID and KID. As predicted by rate-distortion-perception theory, visual quality is less dependent on the bitrate than previous methods.

Vector quantization is a fundamental operation for data compression and vector search. To obtain high accuracy, multi-codebook methods represent each vector using codewords across several codebooks. Residual quantization (RQ) is one such method, which iteratively quantizes the error of the previous step. While the error distribution is dependent on previously-selected codewords, this dependency is not accounted for in conventional RQ as it uses a fixed codebook per quantization step. In this paper, we propose QINCo, a neural RQ variant that constructs specialized codebooks per step that depend on the approximation of the vector from previous steps. Experiments show that QINCo outperforms state-of-the-art methods by a large margin on several datasets and code sizes. For example, QINCo achieves better nearest-neighbor search accuracy using 12-byte codes than the state-of-the-art UNQ using 16 bytes on the BigANN1M and Deep1M datasets.

Large-scale training of latent diffusion models (LDMs) has enabled unprecedented quality in image generation. However, large-scale end-to-end training of these models is computationally costly, and hence most research focuses either on finetuning pretrained models or experiments at smaller scales. In this work we aim to improve the training efficiency and performance of LDMs with the goal of scaling to larger datasets and higher resolutions. We focus our study on two points that are critical for good performance and efficient training: (i) the mechanisms used for semantic level (\eg a text prompt, or class name) and low-level (crop size, random flip, \etc) conditioning of the model, and (ii) pre-training strategies to transfer representations learned on smaller and lower-resolution datasets to larger ones. The main contributions of our work are the following: we present systematic experimental study of these points, we propose a novel conditioning mechanism that disentangles semantic and low-level conditioning, we obtain state-of-the-art performance on CC12M for text-to-image at 512 resolution.

Lossy image compression aims to represent images in as few bits as possible while maintaining fidelity to the original. Theoretical results indicate that optimizing distortion metrics such as PSNR or MS-SSIM necessarily leads to a discrepancy in the statistics of original images from those of reconstructions, in particular at low bitrates, often manifested by the blurring of the compressed images. Previous work has leveraged adversarial discriminators to improve statistical fidelity. Yet these binary discriminators adopted from generative modeling tasks may not be ideal for image compression. In this paper, we introduce a non-binary discriminator that is conditioned on quantized local image representations obtained via VQ-VAE autoencoders. Our evaluations on the CLIC2020, DIV2K and Kodak datasets show that our discriminator is more effective for jointly optimizing distortion (e.g., PSNR) and statistical fidelity (e.g., FID) than the PatchGAN of the state-of-the-art HiFiC model. On CLIC2020, we obtain the same FID as HiFiC with 30-40% fewer bits.

Accelerating MRI scans is one of the principal outstanding problems in the MRI research community. Towards this goal, we hosted the second fastMRI competition targeted towards reconstructing MR images with subsampled k-space data. We provided participants with data from 7,299 clinical brain scans (de-identified via a HIPAA-compliant procedure by NYU Langone Health), holding back the fully-sampled data from 894 of these scans for challenge evaluation purposes. In contrast to the 2019 challenge, we focused our radiologist evaluations on pathological assessment in brain images. We also debuted a new Transfer track that required participants to submit models evaluated on MRI scanners from outside the training set. We received 19 submissions from eight different groups. Results showed one team scoring best in both SSIM scores and qualitative radiologist evaluations. We also performed analysis on alternative metrics to mitigate the effects of background noise and collected feedback from the participants to inform future challenges. Lastly, we identify common failure modes across the submissions, highlighting areas of need for future research in the MRI reconstruction community.

The goal of MRI reconstruction is to restore a high fidelity image from partially observed measurements. This partial view naturally induces reconstruction uncertainty that can only be reduced by acquiring additional measurements. In this paper, we present a novel method for MRI reconstruction that, at inference time, dynamically selects the measurements to take and iteratively refines the prediction in order to best reduce the reconstruction error and, thus, its uncertainty. We validate our method on a large scale knee MRI dataset, as well as on ImageNet. Results show that (1) our system successfully outperforms active acquisition baselines; (2) our uncertainty estimates correlate with error maps; and (3) our ResNet-based architecture surpasses standard pixel-to-pixel models in the task of MRI reconstruction. The proposed method not only shows high-quality reconstructions but also paves the road towards more applicable solutions for accelerating MRI.

In this work, we propose a learning-based variational network (VN) approach for reconstruction of low-dose 3D computed tomography data. We focus on two methods to decrease the radiation dose: (1) x-ray tube current reduction, which reduces the signal-to-noise ratio, and (2) x-ray beam interruption, which undersamples data and results in images with aliasing artifacts. While the learned VN denoises the current-reduced images in the first case, it reconstructs the undersampled data in the second case. Different VNs for denoising and reconstruction are trained on a single clinical 3D abdominal data set. The VNs are compared against state-of-the-art model-based denoising and sparse reconstruction techniques on a different clinical abdominal 3D data set with 4-fold dose reduction. Our results suggest that the proposed VNs enable higher radiation dose reductions and/or increase the image quality for a given dose.

Sparsity-promoting regularization is useful for combining compressed sensing assumptions with parallel MRI for reducing scan time while preserving image quality. Variable splitting algorithms are the current state-of-the-art algorithms for SENSE-type MR image reconstruction with sparsity-promoting regularization. These methods are very general and have been observed to work with almost any regularizer; however, the tuning of associated convergence parameters is a commonly-cited hindrance in their adoption. Conversely, majorize-minimize algorithms based on a single Lipschitz constant have been observed to be slow in shift-variant applications such as SENSE-type MR image reconstruction since the associated Lipschitz constants are loose bounds for the shift-variant behavior. This paper bridges the gap between the Lipschitz constant and the shift-variant aspects of SENSE-type MR imaging by introducing majorizing matrices in the range of the regularizer matrix. The proposed majorize-minimize methods (called BARISTA) converge faster than state-of-the-art variable splitting algorithms when combined with momentum acceleration and adaptive momentum restarting. Furthermore, the tuning parameters associated with the proposed methods are unitless convergence tolerances that are easier to choose than the constraint penalty parameters required by variable splitting algorithms.

Algorithms with Lipschitz bounds such as ISTA and FISTA are useful for solving optimization problems with sparsity-promoting regularizers. However, they can be slow in applications that involve shift-variant system matrices. One example of such an application is MRI with multiple sensitivity coils. We propose a reconstruction algorithm for wavelet regularized SENSE MR image reconstruction that exploits the spatial localization of the wavelet basis and the shift-variant behavior of the MR system matrix to accelerate algorithm convergence. Our results indicate that the proposed method is faster than state-of-the-art variable splitting algorithms in terms of convergence speed for a SENSE-type reconstruction problem even when the variable splitting methods are tuned carefully. Unlike variable splitting methods, the proposed method requires no convergence parameter tuning.