AI Paper Atlas

Research map / Video / Action recognition

Diffusion Action Segmentation

Daochang Liu, Qiyue Li, Anh-Dung Dinh, Tingting Jiang, Mubarak Shah, Chang Xu

ICCV 2023 · Video › Action recognition · approach: Temporal action segmentation · 134 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 position-prior masking during training (ablation)DiffAct · 50SaladsAvg: 86.3 -> 86.8Improved
add all three prior maskings (P, B, R) (ablation)DiffAct · 50SaladsAvg: 86.3 -> 87.4Improved
add boundary alignment loss (ablation)DiffAct · 50SaladsAvg: 86.9 -> 87.4Improved
add temporal smoothness loss (ablation)DiffAct · 50SaladsAvg: 83.6 -> 87.4Improved
boundary loss without smoothness loss (ablation)DiffAct · 50SaladsAcc: 87.0 -> 85.4Hurt
increase inference steps from 1 to 25 (ablation)DiffAct · 50SaladsAvg: 66.6 -> 87.4Improved
increase inference steps beyond 25 to 100 (ablation)DiffAct · 50SaladsAvg: 87.4 -> 87.7Improved
condition on encoder features instead of raw input features (ablation)DiffAct · 50SaladsAvg: 78.7 -> 87.4Improved
choose different encoder layer for conditioning (ablation)DiffAct · 50SaladsAvg: 87.5 / 87.4 / 87.1 for layers 5 / 7 / 9No clear effect
use DiffAct framework with MS-TCN backbone (ablation)MS-TCN · 50SaladsAvg: 77.5 -> 84.0Improved
use diffusion with lightweight decoder instead of ASFormerASFormer · 50SaladsAvg: 81.9 -> 86.6Improved
apply DiffAct on small dataset GTEADiffAct · GTEAEdit/Avg: Avg 88.1 vs SEDT 88.5No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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