Video Corpus Moment Retrieval实验设置记录

2019-arXiv-Temporal Localization of Moments in Video Collections with Natural Language

数据集：DiDeMo, Charades-STA, and ActivityNetcaptions.

评价标准：$R@K$, K ∈ { 1 , 10 , 100 } K\in \{1, 10, 100\} K∈{1,10,100}, I o U ∈ { 0.5 , 0.7 } IoU \in \{0.5, 0.7\} IoU∈{0.5,0.7} & 正确检索的中位数，在两个$IoU$设置下平均（不懂）

性能实验：

2020-ECCV-TVR: A Large-Scale Dataset for Video-Subtitle Moment Retrieval

数据集：TVR

评价标准：性能）$R@K$, K ∈ { 1 , 5 , 10 , 100 } K\in \{1, 5, 10, 100\} K∈{1,5,10,100}, I o U ∈ { 0.5 , 0.7 } IoU \in \{0.5, 0.7\} IoU∈{0.5,0.7} 效率）模型在2080Ti上跑三轮的平均时间（Time spent on data loading, pre-processing, backend model (i.e., ResNet-152, I3D, RoBERTa) feature extraction, etc, is ignored since they should be similar for all methods. We mainly focus on the VCMR task here）

性能实验：

2021-MM-CONQUER: Contextual Query-aware Ranking for Video Corpus Moment Retrieval

数据集：TVR and DiDeMo

评价指标：TVR）$R@K$, K ∈ { 1 , 10 , 100 } K\in \{1, 10, 100\} K∈{1,10,100}, I o U ∈ { 0.5 , 0.7 } IoU \in \{0.5, 0.7\} IoU∈{0.5,0.7} DiDeMo）$R@K$, K ∈ { 1 , 5 , 10 } K\in \{1, 5, 10\} K∈{1,5,10}, I o U ∈ { 0.5 , 0.7 } IoU \in \{0.5, 0.7\} IoU∈{0.5,0.7}

性能实验：

2021-SIGIR-Video Corpus Moment Retrieval with Contrastive Learning

数据集：ActivityNet Captions and TVR

评价指标：性能）$R@K$, K ∈ { 1 , 10 , 100 } K\in \{1, 10, 100\} K∈{1,10,100}, I o U ∈ { 0.5 , 0.7 } IoU \in \{0.5, 0.7\} IoU∈{0.5,0.7} 效率）总体时间和每个查询的平均时间（The time spent on data pre-processing and feature extraction by pre-trained extractor are not counted since the same process applies to all methods）

性能试验：

效率实验：