Motion prediction –forecasting the future trajectories of surrounding vehicles, pedestrians and cyclists is a safety-critical component of the autonomous-driving pipeline that must run in real time on embedded hardware. State-of-the-art predictors, however, are trained on compute clusters and are too large to run on a single consumer GPU, placing both ends of the contemporary pipeline out of reach for an individual researcher. This thesis asks how small a competitive trajectory predictor can be made before its accuracy degrades, and whether the lost accuracy can be recovered through knowledge distillation without enlarging the model or worsening the calibration a downstream planner depends on. The study uses HiVT, a transformer-based Laplace-mixture predictor that is small enough to be both trained and run on a single GPU, evaluated on the Argoverse 1 benchmark. The accuracy–capacity trade-off is first characterised by sweeping the embedding width (128, 64, 32, 16) and locating the point at which a from-scratch student falls measurably below the teacher. The mode-permutation problem is then identified: because HiVT trains its mixture modes with a winner-takes-all loss, the mode slots of two independently trained models do not correspond, so any distillation term that aligns modes by index supervises the student with self-contradictory targets. To resolve this, a permutation-invariant mixture negative-log-likelihood objective is derived that treats the teacher’s modes as an order-free set of soft targets and supports unequal mode counts, with a proof of invariance. Experiments show that a mean-target variant of this objective recovers roughly 84% of the HiVT 32→ HiVT-64 capacity gap (−9.2% minFDE over a matched non-distilled baseline) at zero added inference cost, but degrades full-distribution calibration (mixture NLL +41%, calibration error 5×) by discarding the teacher’s predictive variance. A distribution-matching objective that also distils the teacher’s per-mode scales removes this penalty entirely, leaving the student better calibrated than both the non distilled baseline and the teacher while retaining the full geometric gain. The benefit grows as the student shrinks: at width 16 (55× smaller than the teacher) distribution-matching distillation improves minFDE by −22.7%—roughly 2.5× the width-32 gain—recovering ∼81% of the width-16→width-32 gap, with calibration improving rather than degrading. Distillation thus buys close to a full size-class of accuracy for free, and most where capacity is scarcest. A final efficiency analysis quantifies the deployment frontier: parameter and memory savings are fixed and unconditional (15× at width 32, 55× at width 16), whereas the single-scene latency speed-up is far sublinear and batch-dependent (on CPU ∼3× online, rising to ∼5.5× under modest batching), locating the compression benefit primarily in memory footprint. Overall, the answer to how small a competitive HiVT can be made is encouraging: with a permutation-invariant, calibration-preserving distillation loss, a 55×-smaller student reaches roughly the accuracy of an un-distilled model nearly four times its size at no calibration cost.
Isaac Ding, Daheng Yin, Yili Jin, Rui Qian · 6 authors
User-generated 3D content plays a crucial role in enabling truly immersive experiences within the Web3 metaverse. Although significant progress has been made in dynamic 3D content creation and animation using multi-camera setups, they typically require professional equipment and strict synchronization. In contrast, a dynamic and inclusive Web3 ecosystem is collectively built by digital assets from anyone, at any time, and from anywhere. Most Web3 participants are therefore decentralized and non-professional, and their contributed videos, even of the same scene, are usually captured asynchronously using handheld devices. In this paper, we present GSAlign, a novel framework that achieves high-fidelity dynamic 3D reconstruction from unstructured, user-generated videos in the Web3 metaverse. To address the temporal and spatial misalignments inherent in such data, GSAlign integrates three key modules: Epipolar-guided Temporal Alignment (ETA), Motion-centric Separated Reconstruction (MSR), and Background-guided Global Pose Alignment (BGPA). We detail the design of each module and their integration toward a practical end-to-end implementation. Our evaluation of GSAlign on real-world user-generated videos demonstrates robust reconstruction of dynamic 3D scenes despite unsynchronized captures, sparse views, and handheld camera motion.
GameFi's integration with omnichain non-fungible token initiatives marks a major advance in the gaming and blockchain sectors.This integration brings together the dynamic aspects of decentralised finance (DeFi) with the creative possibilities of non-fungible tokens (NFTs) across numerous blockchain platforms.In the developing industry of GameFi, which combines gaming with financial incentives, non-fungible tokens (NFTs) are used to generate one-of-a-kind in-game assets that may be exchanged, sold, or utilised inside a variety of games and ecosystems.
Abhimanyu Nag, Samrat Gupta, Sudipan Sinha, Arka Datta
Decentralized Finance (DeFi) governance models have become increasingly complex due to the involvement of numerous independent agents, each with their own incentives and strategies. To effectively analyze these systems, we propose using Multi Agent Influence Diagrams (MAIDs) as a powerful tool for modeling and studying the strategic interactions within DeFi governance. MAIDs allow for a comprehensive representation of the decision-making processes of various agents, capturing the influence of their actions on one another and on the overall governance outcomes. In this paper, we study a simple governance game that approximates real governance protocols and compute the Nash equilibria using MAIDs. We further outline the structure of a MAID in MakerDAO.