Predicting Event Memorability from Contextual Visual Semantics

Authors: Qianli Xu, Fen Fang, Ana Molino, Vigneshwaran Subbaraju, Joo-Hwee Lim

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this study, we investigate factors that affect event memorability according to a cued recall process. Specifically, we explore whether event memorability is contingent on the event context, as well as the intrinsic visual attributes of image cues. We design a novel experiment protocol and conduct a large-scale experiment with 47 elder subjects over 3 months. Subjects memory of life events is tested in a cued recall process. Using advanced visual analytics methods, we build a first-of-its-kind event memorability dataset (called R3) with rich information about event context and visual semantic features. Furthermore, we propose a contextual event memory network (CEMNet) that tackles multi-modal input to predict item-wise event memorability, which outperforms competitive benchmarks.
Researcher Affiliation Collaboration Qianli Xu Institute for Infocomm Research A*STAR, Singapore qxu@i2r.a-star.edu.sg Fen Fang Institute for Infocomm Research A*STAR, Singapore fang_fen@i2r.a-star.edu.sg Ana Garcia del Molino Byte Dance AI Lab Singapore a.g.delmolino@gmail.com Vigneshwaran Subbaraju Institute of High Performance Computing A*STAR, Singapore vigneshwaran_subbaraju@ihpc.a-star.edu.sg Joo Hwee Lim Institute for Infocomm Research, A*STAR Nanyang Technological University, Singapore joohwee@i2r.a-star.edu.sg
Pseudocode No The paper provides a model architecture diagram (Figure 2) and describes the components, but it does not include any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes Source code is available at https: //github.com/ffzzy840304/Predicting-Event-Memorability.
Open Datasets No The paper mentions creating and using the 'R3' dataset for its experiments, stating it is the 'first, large-scale dataset that includes rich visual and contextual information associated with event memory from the real-world context'. However, it does not provide any specific link, DOI, repository, or explicit statement of public availability for this dataset.
Dataset Splits Yes First, we randomly split the dataset with a 4:1 train-test ratio based on subject index. ... We perform 5 random splits and evaluate the average performance (i.e., 5-fold cross-validation).
Hardware Specification Yes Our models are trained (with Adam optimizer) and evaluated on Pytorch platform using a machine with NVIDIA Ge Force GTX 1080 GPU.
Software Dependencies No The paper mentions 'Pytorch platform' but does not specify its version or any other software dependencies with their specific version numbers (e.g., 'Python 3.x', 'CUDA x.x', or specific library versions).
Experiment Setup Yes For MLP, AMNet and ICNet, the batch size are set as 64, 16 and 32 respectively; learning rate are set as 0.001, 0.0001 and 0.0001; training epochs are set as 100.