Semi-supervised Open-World Object Detection
Authors: Sahal Shaji Mullappilly, Abhishek Singh Gehlot, Rao Muhammad Anwer , Fahad Shahbaz Khan, Hisham Cholakkal
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments on 4 datasets including MS COCO, PASCAL, Objects365 and DOTA demonstrate the effectiveness of our approach. |
| Researcher Affiliation | Academia | Sahal Shaji Mullappilly1, Abhishek Singh Gehlot1, Rao Muhammad Anwer1, Fahad Shahbaz Khan1,2, Hisham Cholakkal1 1Mohamed bin Zayed University of Artificial Intelligence 2Link oping University |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our source code, models and splits are available here https://github.com/sahalshajim/SS-OWFormer. |
| Open Datasets | Yes | Datasets: We evaluate our SS-OWOD framework on MSCOCO (Lin et al. 2014), Pascal VOC (Everingham et al. 2010), DOTA (Xia et al. 2018) and Objects365 (Shao et al. 2019) for OWOD problem. |
| Dataset Splits | No | The paper mentions 'a held-out validation set' in section 4.1 but clarifies that the ORE's energy-based unknown identifier (EBUI) relies on it and they omit it. It does not provide explicit details about validation splits used in their own experiments (e.g., specific percentages or sample counts for validation data). |
| Hardware Specification | No | The paper mentions 'The computational resources were provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by the Swedish Research Council through grant agreement no. 2022-06725, and by the Berzelius resource, provided by the Knut and Alice Wallenberg Foundation at the National Supercomputer Centre.' but does not specify any exact hardware details like GPU or CPU models. |
| Software Dependencies | No | The paper mentions 'The transformer architecture is a version of Deformable DETR', 'Multi-scale feature maps are taken from Image Net pre-trained Res Net50', and 'Training is carried out for 50 epochs using ADAM optimizer (Kingma and Ba 2014) with weight decay (Adam W)'. While it names software components, it does not provide specific version numbers for them or any underlying libraries/frameworks. |
| Experiment Setup | Yes | Number of queries is set to M = 250 to account for the high number of instances in satellite images, while the threshold for the selection of pseudo-labels is set to top-10. Training is carried out for 50 epochs using ADAM optimizer (Kingma and Ba 2014) with weight decay (Adam W) and learning rate set to 10 4. |