Classification by Attention: Scene Graph Classification with Prior Knowledge

Authors: Sahand Sharifzadeh, Sina Moayed Baharlou, Volker Tresp5025-5033

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

Reproducibility Variable Result LLM Response
Research Type Experimental We train our models on the common split of Visual Genome (Krishna et al. 2017) dataset... We report the experimental results on the test set... Table 1 compares the performance of our model to the state-of-the-art... Figure 4 shows our ablation study...
Researcher Affiliation Collaboration Sahand Sharifzadeh,1 Sina Moayed Baharlou,1* Volker Tresp1,2 1Ludwig Maximilian University of Munich 2Siemens AG
Pseudocode No The paper describes its methods using mathematical equations and prose but does not include a clearly labeled “Pseudocode” or “Algorithm” block.
Open Source Code Yes For those results, please refer to our official code repository. The splits are available at: https://github.com/sharifza/schemata
Open Datasets Yes We train our models on the common split of Visual Genome (Krishna et al. 2017) dataset containing images labeled with their scene graphs (Xu et al. 2017).
Dataset Splits Yes We train our models on the common split of Visual Genome (Krishna et al. 2017) dataset... We report the experimental results on the test set... we uniformly sample two splits with 1% and 10% of VG. The images in each split are considered as labeled.
Hardware Specification No The paper mentions neural network architectures like VGG16 and ResNet-50 as backbones but does not provide specific details about the hardware (e.g., GPU models, CPU, memory) used for running the experiments.
Software Dependencies No The paper mentions various models and architectures (e.g., VGG16, ResNet-50, BYOL, GCNs, LSTMs) and refers to the Visual Genome dataset, but it does not specify version numbers for any software dependencies like deep learning frameworks (e.g., PyTorch, TensorFlow), programming languages (e.g., Python), or other libraries.
Experiment Setup No The paper describes general training strategies like using categorical cross-entropy loss, teacher forcing, and Leaky ReLU non-linearities. It mentions training for '8 assimilations' and using Adam. However, it does not provide specific hyperparameter values such as learning rate, batch size, or the total number of training epochs.