Reproduce “Concepts in Motion: Temporal Concept Bottleneck Model for Interpretable Video Classification” 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
Patrick Knab, Sascha Marton, Philipp J. Schubert, Drago Guggiana, Christian Bartelt
MoTIF models video actions through human-readable concepts tracked across temporal windows. A vision-language model proposes object and action concepts from training videos, and per-concept temporal attention connects each concept’s activations across time without mixing concept channels. On Breakfast with the PE-G/14 backbone, it reaches 87.5 ± 4.9% versus 75.8 ± 7.1% for Global CBM. This comparison does not establish parity with the strongest black-box models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.