Reproduce “LVMT: Video Mask Transformer for Long-term Video Segmentation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Narges Norouzi, Niccolò Cavagnero, Idil Esen Zulfikar, Bastian Leibe, Gijs Dubbelman, Daan de Geus
LVMT tracks object identities during online video segmentation. LVMT uses a GRU to update each object’s compact query memory and truncated query propagation to train longer clips without backpropagating through the entire clip. On OVIS validation, LVMT reaches 56.4 ± 0.4 AP versus PMT’s 51.8 ± 0.3 at similar throughput, 95 versus 97 FPS; training tests extend to 20 frames.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.