The rapidly evolving landscape of products, surfaces, policies, and regulations poses significant challenges for deploying state-of-the-art recommendation models at industry scale, primarily due to data fragmentation across domains and escalating infrastructure costs that hinder sustained quality improvements.

Recent advances in recommendation scaling laws have led to foundation models of unprecedented complexity. While these models offer superior performance, their computational demands make real-time serving impractical, often forcing practitioners to rely on knowledge distillation—compromising serving quality for efficiency. To address this challenge, we present SOLARIS (Speculative Offloading of Latent-bAsed Representation for Inference Scaling), a novel framework inspired by speculative decoding. SOLARIS proactively precomputes user-item interaction embeddings by predicting which user-item pairs are likely to appear in future requests, and asynchronously generating their foundation model representations ahead of time. This approach decouples the costly foundation model inference from the latency-critical serving path, enabling real-time knowledge transfer from models previously considered too expensive for online use. Deployed across Meta's advertising system serving billions of daily requests, SOLARIS achieves 0.67% revenue-driving top-line metrics gain, demonstrating its effectiveness at scale.

We propose a method for unsupervised video object segmentation by transferring the knowledge encapsulated in image-based instance embedding networks. The instance embedding network produces an embedding vector for each pixel that enables identifying all pixels belonging to the same object. Though trained on static images, the instance embeddings are stable over consecutive video frames, which allows us to link objects together over time. Thus, we adapt the instance networks trained on static images to video object segmentation and incorporate the embeddings with objectness and optical flow features, without model retraining or online fine-tuning. The proposed method outperforms state-of-the-art unsupervised segmentation methods in the DAVIS dataset and the FBMS dataset.

A large-scale video quality dataset called the VideoSet has been constructed recently to measure human subjective experience of H.264 coded video in terms of the just-noticeable-difference (JND). It measures the first three JND points of 5-second video of resolution 1080p, 720p, 540p and 360p. Based on the VideoSet, we propose a method to predict the satisfied-user-ratio (SUR) curves using a machine learning framework. First, we partition a video clip into local spatial-temporal segments and evaluate the quality of each segment using the VMAF quality index. Then, we aggregate these local VMAF measures to derive a global one. Finally, the masking effect is incorporated and the support vector regression (SVR) is used to predict the SUR curves, from which the JND points can be derived. Experimental results are given to demonstrate the performance of the proposed SUR prediction method.

With the development of Fully Convolutional Neural Network (FCN), there have been progressive advances in the field of semantic segmentation in recent years. The FCN-based solutions are able to summarize features across training images and generate matching templates for the desired object classes, yet they overlook intra-class difference (ICD) among multiple instances in the same class. In this work, we present a novel fine-to-coarse learning (FCL) procedure, which first guides the network with designed 'finer' sub-class labels, whose decisions are mapped to the original 'coarse' object category through end-to-end learning. A sub-class labeling strategy is designed with unsupervised clustering upon deep convolutional features, and the proposed FCL procedure enables a balance between the fine-scale (i.e. sub-class) and the coarse-scale (i.e. class) knowledge. We conduct extensive experiments on several popular datasets, including PASCAL VOC, Context, Person-Part and NYUDepth-v2 to demonstrate the advantage of learning finer sub-classes and the potential to guide the learning of deep networks with unsupervised clustering.

A contour-like artifact called false contour is often observed in large smooth areas of decoded images and video. Without loss of generality, we focus on detection and removal of false contours resulting from the state-of-the-art High Efficiency Video Coding codec. First, we identify the cause of false contours by explaining the human perceptual experiences on them with specific experiments. Next, we propose a precise pixel-based false contour detection method based on the evolution of a false contour candidate (FCC) map. The number of points in the FCC map becomes fewer by imposing more constraints step by step. Special attention is paid to separating false contours from real contours such as edges and textures in the video source. Then, a decontour method is designed to remove false contours in the exact contour position while preserving edge/texture details. Extensive experimental results are provided to demonstrate the superior performance of the proposed false contour detection and removal method in both compressed images and videos.

Evaluation of coding efficiency is traditionally modeled as a continuous rate-distortion (R-D) function, where the peak signal-to-noise ratio (PSNR) is adopted as the quality measure. Although the PSNR-versus-bitrate curve offers some useful tradeoff information between video quality and coding bit-rates, it does not take human perceptual experience into account. In this work, by following the recent image/video quality assessment framework based on the just-noticeable-difference (JND) notion, we conduct a subjective test for HEVC (High Efficiency Video Codec) video to measure the QP value that lies in the boundary of perceptually lossless and lossy coded bit streams for each human subject. This is also known as the first JND point. It is observed that the statistics of the first JND points of 30 subjects follows the normal distribution for a great majority of test sequences. Finally, a machine-learning approach is proposed to predict the mean of the group-based JND distribution based on extracted video features. It is shown by experimental results that the mean JND point can be predicted accurately.

A two stages car detection method using deformable part models with composite feature sets (DPM/CF) is proposed to recognize cars of various types and from multiple viewing angles. In the first stage, a HOG template is matched to detect the bounding box of the entire car of a certain type and viewed from a certain angle (called a t/a pair), which yields a region of interest (ROI). In the second stage, various part detectors using either HOG or the convolution neural network (CNN) features are applied to the ROI for validation. An optimization procedure based on latent logistic regression is adopted to select the optimal part detector's location, window size, and feature to use. Extensive experimental results indicate the proposed DPM/CF system can strike a balance between detection accuracy and training complexity.