PairSelect
Token compression for weakly supervised video anomaly detection
ICASSP 2027 · Submitted
Reduce long-video encoding cost while retaining original token vectors and temporal coverage.
- My role
- First author
- Methods & tools
- PyTorch · VideoMAEv2 · VideoMAE · TimeSformer
01
Problem & challenge
Continuous long-video encoding is expensive. Compression must preserve temporal coverage and work across different visual backbones.
02
Key contributions
My contribution
Independently developed local pairing, norm ranking and group quotas after Transformer block 6. Adapted four detector systems without retraining for compression or budget changes.
Method
Contextualize all tokens through the first six dense Transformer blocks.
Apply local pairing, norm ranking and group quotas, preserving original vectors and temporal coverage.
Run the remaining blocks with fewer tokens and reuse detector weights, without retraining for compression or budget changes.
03
Experimental results
On UCF-Crime / XD-Violence at batch size 32, 40% token deletion yields 1.22–1.28× encoder speedup. VideoMAEv2 UCF AUC rises from 80.63% to 81.02%; TimeSformer XD PR-AUC rises from 76.69% to 77.37%. At 60% token deletion, the encoder reaches up to 1.47× encoder speedup.
Research in progress; code and manuscript are not public.