AIbase

ARDA-VSLNet-FiLM-Enhanced-Knowledge-Distillation-for-Efficient-NLQ-Localization-in-Ego4D-Videos

Public

Efficient NLVL on Ego4D. Benchmark VSLBase/VSLNet with BERT/GloVE, EgoVLP/Omnivore, configurable FiLM conditioning layer (Perez et al., FiLM: Visual Reasoning with a General Conditioning Layer), explores compression via KD, including CBKD (Lan et al., Counterclockwise Block‑by‑Block Knowledge Distillation), prototype post‑training quantization.

Creat2025-04-01T20:47:05
Update2025-07-27T00:58:56
0
Stars
0
Stars Increase

Related projects