Detecting Out-Of-Context Objects Using Graph Contextual Reasoning Network
Authors: Manoj Acharya, Anirban Roy, Kaushik Koneripalli, Susmit Jha, Christopher Kanan, Ajay Divakaran
IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In order to evaluate our approach, we create a large-scale dataset by adding OOC object instances to the COCO images. We also evaluate on recent OCD benchmark. Our results show that GCRN outperforms competitive baselines in detecting OOC objects and correctly detecting in-context objects. |
| Researcher Affiliation | Collaboration | Manoj Acharya1,2 , Anirban Roy1 , Kaushik Koneripalli1 , Susmit Jha1 , Christopher Kanan2 and Ajay Divakaran1 1SRI International, Menlo Park CA 94025, USA 2Rochester Institute of Technology, Rochester NY 14623, USA |
| Pseudocode | No | The paper describes the model architecture and learning process in text and equations but does not include any pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code and data: https://nusci.csl.sri.com/project/trinity-ooc |
| Open Datasets | Yes | The COCO-OOC Dataset. To evaluate the performance of OOC detection, we create a large-scale OOC dataset based on the COCO object detection benchmark [Lin et al., 2014]. ... The OCD Dataset. In addition to our COCO-OOC dataset, we also use the recent OCD benchmark with synthetically generated OOC indoor scenes [Bomatter et al., 2021]. |
| Dataset Splits | No | The paper mentions training on 'COCO train set' and testing on 'OOC images' from COCO-OOC and OCD, but does not provide specific percentages or counts for training, validation, and test splits needed to reproduce the data partitioning. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models or memory specifications used for running the experiments. |
| Software Dependencies | No | The paper mentions using the 'DGL toolbox', 'Mask RCNN' and 'Res Net50' but does not specify their version numbers. |
| Experiment Setup | Yes | For both Rep G and Con G, we consider a GCN with four graph convolution layers with residual connections between the layers. The numbers of neurons at these layers are 256, 128, 64, and 64 respectively. GCNs are trained using an Adam W optimizer with a learning rate of 0.001 without decay. For our GCRN framework, we first train Rep G for five epochs in the first phase and alternate between Rep G and Con G until convergence. In our experiments, convergence is reached within ten iterations. |