AI Paper Atlas

Research map / Video / Action recognition

RAP: Retrieval-Augmented Planner for Adaptive Procedure Planning in Instructional Videos

Ali Zare, Yulei Niu, Hammad Ayyubi, Shih-Fu Chang

ECCV 2024 · Video › Action recognition · approach: Action anticipation and latent actions · 5 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add retrieval-augmented memory to the base planner (T=3) (ablation)RAP · CrossTask T=3SR: 25.42 -> 29.23Improved
add retrieval memory in variable-length setting (ablation)RAP · CrossTask variable lengthSR: 18.54 -> 20.28Improved
add retrieval memory on COIN variable length (ablation)RAP · COIN variable lengthSR: 17.86 -> 19.14Improved
add weakly-supervised pseudo-labelled unannotated videos (100%) (ablation)RAP · CrossTask variable lengthSR: 20.28 -> 21.79Improved
use only 25% of unannotated data (ablation)RAP · CrossTask variable lengthSR: 20.28 -> 19.20Hurt
add unannotated data at T=4RAP · CrossTask T=4mIoU: 71.49 -> 71.33No clear effect
use explicit retrieval memory instead of implicit memory (ablation)RAP · CrossTask T=3SR: 26.78 -> 29.23Improved
use END token instead of separate length predictor (ablation)RAP · CrossTask variable lengthplan-length accuracy (%): 21.32 -> 53.47Improved
retrieval head alone vs base planner (ablation)RAP · CrossTask variable lengthSR: 18.54 -> 11.42Hurt
increase datastore size from 1000 to 1500 (ablation)RAP · CrossTask variable lengthSR: 20.28 -> 18.03Hurt
change lambda from 0.1 to 0.05 or 0.15 (ablation)RAP · CrossTask variable lengthSR: 20.28 -> 19.86No clear effect
RAP vs Skip-Plan/PDPP on COIN T=4 mAccRAP · COIN T=4mAcc: 44.10 vs 42.11Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy