Hao Shi

Transforming point seeds into temporally consistent object instances, PS-Track sets a new benchmark in multi-object tracking without relying on traditional bounding box annotations.

Achieving state-of-the-art kanji reading accuracy, Sarashina2.2-TTS sets a new standard for Japanese speech synthesis while ensuring cross-lingual robustness.

SLMs show improved dialectal understanding when trained on dialectal data, revealing a critical pathway for enhancing performance in diverse linguistic contexts.

VL2Spike boosts Spikformer performance by over 6% while slashing energy consumption to just 15.7% of traditional methods.

ReasonAlloc reallocates KV cache resources in real-time, achieving superior reasoning efficiency with minimal overhead.

MemoryVLA++ achieves up to 28% performance gains in robotic manipulation tasks by integrating memory and imagination, transforming how robots handle temporal dependencies.

Even simple event-based augmentation can rescue vision-language-action models from complete failure in low-light or high-blur scenarios, boosting task success from 0% to 90% in some cases.

Achieve state-of-the-art 3D occupancy prediction by unifying geometric, semantic, and textual information within a novel omnidirectional framework.