REFINE: Prediction Fusion Network for Panoptic Segmentation
Authors: Jiawei Ren, Cunjun Yu, Zhongang Cai, Mingyuan Zhang, Chongsong Chen, Haiyu Zhao, Shuai Yi, Hongsheng Li2477-2485
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we evaluate our approach on COCO dataset (Lin et al. 2014) for panoptic segmentation. More experiments (including those on Cityscapes (Cordts et al. 2016)) can be found in the supplementary material due to space constraint. Comparisons with state-of-the-art methods demonstrate the effectiveness of our overall framework. We also evaluate each sub-module of our method together with a detailed analysis. We further decompose our method to find out the contribution of each component. Qualitative results are shown in Figure 2. |
| Researcher Affiliation | Collaboration | Sense Time Research Nanyang Technological University Multimedia Laboratory, The Chinese University of Hong Kong |
| Pseudocode | Yes | The pseudo-code for UE can be found in the supplementary material. |
| Open Source Code | No | The paper does not contain an explicit statement or link indicating that the source code for their methodology is publicly available. |
| Open Datasets | Yes | In this section, we evaluate our approach on COCO dataset (Lin et al. 2014) for panoptic segmentation. More experiments (including those on Cityscapes (Cordts et al. 2016)) can be found in the supplementary material due to space constraint. |
| Dataset Splits | Yes | We take the default 118k/5k/20k split for train/val/test from COCO2017. |
| Hardware Specification | No | The paper does not specify any particular hardware details such as GPU models, CPU models, or memory specifications used for experiments. |
| Software Dependencies | Yes | Cascade Panoptic-FPN1, which adds deformable convolutions and Cascade-RCNN (Cai and Vasconcelos 2018) to the original Panoptic-FPN in order to obtain better bounding box localization and mask prediction. 1implemented in detectron2 with its version at commit 8999946 |
| Experiment Setup | Yes | Cascade Panoptic FPN is trained for an even longer schedule, 270k iterations while we train our model for 180k iterations under the same setup. |