Deep Variational Instance Segmentation
Authors: Jialin Yuan, Chao Chen, Fuxin Li
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on PASCAL VOC 2012 and the MSCOCO 2017 dataset show that the proposed approach efficiently tackles the instance segmentation task. |
| Researcher Affiliation | Academia | Jialin Yuan Co RIS Institute Oregon State University yuanjial@oregonstate.edu Chao Chen Stony Brook University chao.chen.1@stonybrook.edu Li Fuxin Co RIS Institute Oregon State University lif@oregonstate.edu |
| Pseudocode | No | The paper does not contain any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | The source code and trained models are released at https://github.com/jia2lin3yuan1/2020-instanceSeg. |
| Open Datasets | Yes | PASCAL VOC 2012 consists of 20 object classes and one background class. It has been the benchmark challenge for segmentation over the years. The original dataset contains 1,464, 1,449, and 1,456 images for training, validation, and testing. It is augmented by extra annotations from [16], resulting in 10,582 training images. COCO is a very challenging dataset for instance segmentation and object detection. It has 115,000 images and 5,000 images for training and validation, respectively. 20,000 images are used as test-dev from the split of 2017. There are 80 instance classes for instance segmentation and object detection challenge. We train our model on the train 2017 subset and run prediction on val 2017 and test-dev 2017 subsets respectively. We adopt the public cocoapi to report the performance metrics AP, AP50, AP75, APS, APM, and APL. |
| Dataset Splits | Yes | PASCAL VOC 2012... The original dataset contains 1,464, 1,449, and 1,456 images for training, validation, and testing. COCO... It has 115,000 images and 5,000 images for training and validation, respectively. 20,000 images are used as test-dev from the split of 2017. We train our model on the train 2017 subset and run prediction on val 2017 and test-dev 2017 subsets respectively. |
| Hardware Specification | Yes | Moreover, in Table 4, speed analysis on a V100 GPU (all post-processing included) is shown in the column FPS. |
| Software Dependencies | No | The paper mentions various models and components (e.g., ResNet, FPN, DeepLab-v3, Huber loss, cocoapi) but does not provide specific version numbers for any of the software dependencies used in the experiments. |
| Experiment Setup | Yes | We scale the input image to 513 513 for PASCAL and with the minimal edge equal to 700 for COCO (preserving the height-to-width ratio). The window size for computing relative loss is set to 128 throughout all experiments... And we initialize the backbone network with the pre-trained weights for the semantic segmentation task on PASCAL and the pre-trained weights for the object detection task on COCO. The base setting we use is w = 129, c = 8, r = 8. In experiments, we set m1 = 2. We set m2 = 1 in practice. We set θ = 0.1 throughout the work. After we obtain the real-valued instance labels, we apply the mean-shift segmentation algorithm on it with different bandwidthes, 0.9 and 0.4 to discretize it to two different label maps. |