EmbedMask: Embedding Coupling for Instance Segmentation
Authors: Hui Ying, Zhaojin Huang, Shu Liu, Tianjia Shao, Kun Zhou
IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Current instance segmentation methods can be categorized into segmentation-based methods and proposal-based methods. ... Embed Mask outperforms the state-of-the-art instance segmentation method Mask R-CNN on the challenging COCO dataset, obtaining more detailed masks at a higher speed. ... 4 Experiments 4.1 Experimental Settings We follow the settings of FCOS [Tian et al., 2019b] in our experiments, which chooses the large-scale detection benchmark COCO, and uses the COCO trainval35k split (115K images) for training, minival split (5K images) for ablation study and test-dev (20K images) for reporting the main results. |
| Researcher Affiliation | Collaboration | 1State Key Lab of CAD&CG, Zhejiang University 2Kuai Shou 3Smart More |
| Pseudocode | No | The paper describes the method using text, figures (network architecture diagrams), and equations, but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing its own source code or a direct link to a repository containing its implementation. |
| Open Datasets | Yes | We follow the settings of FCOS [Tian et al., 2019b] in our experiments, which chooses the large-scale detection benchmark COCO, and uses the COCO trainval35k split (115K images) for training, minival split (5K images) for ablation study and test-dev (20K images) for reporting the main results. ... COCO dataset [Lin et al., 2014] |
| Dataset Splits | Yes | We follow the settings of FCOS [Tian et al., 2019b] in our experiments, which chooses the large-scale detection benchmark COCO, and uses the COCO trainval35k split (115K images) for training, minival split (5K images) for ablation study and test-dev (20K images) for reporting the main results. |
| Hardware Specification | Yes | on an NVIDIA Ge Force 2080 Ti GPU |
| Software Dependencies | No | The paper mentions frameworks and optimizers like FCOS, maskrcnn-benchmark, Adelai Det, and SGD, but does not provide specific version numbers for these or other software dependencies such as Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | Unless noted, the input images are resized with the shorter side being 800 while the longer side being no longer than 1333. ... We train all the models with SGD using an initial learning rate of 0.01 and batch size of 16, with constant warm-up of 500 iterations. ... The models are trained for 12 epochs (90k iterations) by default, but more epochs are applied when training with the Res Net-101 backbone. ... In the main results, we set embedding dim D = 16. ... λ1 = 0.5 by default. ... We find that when the threshold is 0.52 or 0.53, the mask AP is the highest. Hence we use 0.52 for the Res Net-50 backbone and 0.53 for the Res Net-101 backbone. |