Self-supervised Amodal Video Object Segmentation
Authors: Jian Yao, Yuxin Hong, Chiyu Wang, Tianjun Xiao, Tong He, Francesco Locatello, David P Wipf, Yanwei Fu, Zheng Zhang
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The proposed framework achieves the state-of-the-art performance on the synthetic amodal segmentation benchmark FISHBOWL and the real world benchmark KINS-Video-Car. Further, it lends itself well to being transferred to novel distributions using test-time adaptation, outperforming existing models even after the transfer to a new distribution. The experiment code is available at https://github.com/amazon-science/self-supervisedamodal-video-object-segmentation. 4 Experiments We evaluate the proposed pipeline on both close-world setting with no distribution shifts between training set and test set, as well as the setting has distribution shifts with new object types. |
| Researcher Affiliation | Collaboration | 1 School of Management, Fudan University 2 School of Data Science, Fudan University 3 University of California, Berkeley 4 Amazon Web Services |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The experiment code is available at https://github.com/amazon-science/self-supervisedamodal-video-object-segmentation. |
| Open Datasets | Yes | This dataset [33] consists of 10,000 training and 1,000 validation and test videos recorded from a publicly available Web GL demo of an aquarium [1]... KINS is an image-level amodal dataset labeled from the city driving dataset KITTI [9]. In order to have Sa Vos work with KINS, we match images in KINS to its original video frame in KITTI. |
| Dataset Splits | Yes | This dataset [33] consists of 10,000 training and 1,000 validation and test videos... we re-split the original KITTI training set into three subsets for training, validation and test. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper mentions using CNNs and LSTMs for implementation but does not specify software dependencies with version numbers (e.g., PyTorch 1.9, TensorFlow 2.x, CUDA 11.x). |
| Experiment Setup | No | The paper mentions general aspects of the experimental setup, such as evaluation metrics and stopping criteria for test-time adaptation ("stop the optimization if the visible-part Io U improves less than 0.01 in the last 10 iterations"), but it lacks specific hyperparameters (e.g., learning rate, batch size, number of epochs for main training) or detailed configuration settings for the models used. |