Weakly Supervised Instance Segmentation using the Bounding Box Tightness Prior
Authors: Cheng-Chun Hsu, Kuang-Jui Hsu, Chung-Chi Tsai, Yen-Yu Lin, Yung-Yu Chuang
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experimental results show that our method performs favorably against existing weakly supervised methods and even surpasses some fully supervised methods for instance segmentation on the PASCAL VOC dataset. The proposed method is evaluated in this section. First, we describe the adopted dataset and evaluation metrics. Then, the performance of the proposed method and the competing methods is compared and analyzed. Finally, the ablation studies on each proposed component and several baseline variants are conducted. |
| Researcher Affiliation | Collaboration | Cheng-Chun Hsu1 Kuang-Jui Hsu2 Chung-Chi Tsai2 Yen-Yu Lin1,3 Yung-Yu Chuang1,4 1Academia Sinica 2Qualcomm Technologies, Inc. 3National Chiao Tung University 4National Taiwan University |
| Pseudocode | No | The paper does not contain any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github.com/chengchunhsu/WSIS_BBTP. |
| Open Datasets | Yes | The Pascal VOC 2012 [22] dataset is widely used in the literature of instance segmentation [1, 17, 61]. This dataset consists of 20 object classes. Following the previous work [17], we use the augmented Pascal VOC 2012 dataset [71] which contains totally 10, 582 training images. In addition, for fair comparison with the SDI method [17], we also train our method using the additional training images from the MS COCO [72] dataset and report the performance. |
| Dataset Splits | No | The paper mentions using training images but does not specify a validation split or training/validation/test split percentages/counts. |
| Hardware Specification | Yes | During training, the network is optimized on a machine with four Ge Force GTX 1080 Ti GPUs. |
| Software Dependencies | No | The paper states 'We implement the proposed method using Py Torch.' but does not provide specific version numbers for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | The batch size, learning rate, weight decay, momentum, and the number of the iterations are set to 8, 10 2, 10 4, 0.9 and 22k, respectively. We choose ADAM [70] as the optimization solver because of its fast convergence. For data augmentation, following the setting used in Mask R-CNN, we horizontally flip each image with probability 0.5, and randomly resize each image so that the shorter side is larger than 800 pixels and the longer side is smaller than 1, 333 pixels, while maintaining the original aspect ratio. |