MBZUAI researchers reduce video object tracking to two decoding rounds
Parallel Tube Decoding completed a trajectory faster than sequential generation in a VidSTG test.
Event / data period:
Published: · Updated:
Researchers at MBZUAI and UC Merced developed Parallel Tube Decoding (PTD), a method for locating objects and events in video over time. It aims to generate an entire object trajectory in two decoding rounds: first identifying the relevant event interval, then predicting the bounding boxes across that interval together.
In a controlled VidSTG comparison described by MBZUAI on 6 October, PTD completed a trajectory in 0.4 seconds, compared with 31.6 seconds for standard autoregressive generation. Box throughput in that test was 45.9 boxes per second for PTD and 0.5 for the baseline. These are research results on a specific video-grounding benchmark, not a guarantee of general-use or production performance.
Data
VidSTG controlled comparison
Video-grounding model tested in a controlled comparison
- · Occurred
Research findings published
The study reports model results under the synthetic or controlled benchmark conditions stated in the source.
Methodology
Measurements apply to the models, baseline method and VidSTG benchmark described in the article. They are not universal speed figures and are not aggregated across countries or models.
Updated:
Subjects and places
Subject proposals
- Science and technologySuggested
- Preserved editorial category/topic; broad candidate onlymbzuai.ac.ae ↗
Geographic references
Only source-provided geographic references are shown. Opening an external map is optional. Optional story maps use OpenStreetMap. Opening a map sends your IP address and this site’s address to the map provider; location data remains readable as text without loading the map.
- United Arab Emirates · Place role: Coverage · Place type: Country or area · Location precision: RegionalEvidence
- Explicit preserved editorial country/region; no city or national centroid inferredmbzuai.ac.ae ↗